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_pager.h> 58 #include <vm/swap_pager.h> 59 60 #include <sys/buf2.h> 61 #include <vm/vm_page2.h> 62 63 #include <vfs/fifofs/fifo.h> 64 #include <vfs/tmpfs/tmpfs_vnops.h> 65 #if 0 66 #include <vfs/tmpfs/tmpfs.h> 67 #endif 68 #include "tmpfs.h" 69 70 static void tmpfs_strategy_done(struct bio *bio); 71 72 static __inline 73 void 74 tmpfs_knote(struct vnode *vp, int flags) 75 { 76 if (flags) 77 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 78 } 79 80 81 /* --------------------------------------------------------------------- */ 82 83 static int 84 tmpfs_nresolve(struct vop_nresolve_args *v) 85 { 86 struct vnode *dvp = v->a_dvp; 87 struct vnode *vp = NULL; 88 struct namecache *ncp = v->a_nch->ncp; 89 struct tmpfs_node *tnode; 90 struct mount *mp; 91 92 int error; 93 struct tmpfs_dirent *de; 94 struct tmpfs_node *dnode; 95 96 mp = dvp->v_mount; 97 lwkt_gettoken(&mp->mnt_token); 98 99 dnode = VP_TO_TMPFS_DIR(dvp); 100 101 de = tmpfs_dir_lookup(dnode, NULL, ncp); 102 if (de == NULL) { 103 error = ENOENT; 104 } else { 105 /* 106 * Allocate a vnode for the node we found. 107 */ 108 tnode = de->td_node; 109 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 110 LK_EXCLUSIVE | LK_RETRY, &vp); 111 if (error) 112 goto out; 113 KKASSERT(vp); 114 } 115 116 out: 117 /* 118 * Store the result of this lookup in the cache. Avoid this if the 119 * request was for creation, as it does not improve timings on 120 * emprical tests. 121 */ 122 if (vp) { 123 vn_unlock(vp); 124 cache_setvp(v->a_nch, vp); 125 vrele(vp); 126 } else if (error == ENOENT) { 127 cache_setvp(v->a_nch, NULL); 128 } 129 130 lwkt_reltoken(&mp->mnt_token); 131 return (error); 132 } 133 134 static int 135 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v) 136 { 137 struct vnode *dvp = v->a_dvp; 138 struct vnode **vpp = v->a_vpp; 139 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp); 140 struct ucred *cred = v->a_cred; 141 struct mount *mp; 142 int error; 143 144 *vpp = NULL; 145 146 mp = dvp->v_mount; 147 lwkt_gettoken(&mp->mnt_token); 148 149 /* Check accessibility of requested node as a first step. */ 150 error = VOP_ACCESS(dvp, VEXEC, cred); 151 if (error != 0) { 152 lwkt_reltoken(&mp->mnt_token); 153 return error; 154 } 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 } 164 165 lwkt_reltoken(&mp->mnt_token); 166 167 return (*vpp == NULL) ? ENOENT : 0; 168 } 169 170 /* --------------------------------------------------------------------- */ 171 172 static int 173 tmpfs_ncreate(struct vop_ncreate_args *v) 174 { 175 struct vnode *dvp = v->a_dvp; 176 struct vnode **vpp = v->a_vpp; 177 struct namecache *ncp = v->a_nch->ncp; 178 struct vattr *vap = v->a_vap; 179 struct ucred *cred = v->a_cred; 180 struct mount *mp; 181 int error; 182 183 mp = dvp->v_mount; 184 lwkt_gettoken(&mp->mnt_token); 185 186 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK); 187 188 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 189 if (error == 0) { 190 cache_setunresolved(v->a_nch); 191 cache_setvp(v->a_nch, *vpp); 192 tmpfs_knote(dvp, NOTE_WRITE); 193 } 194 195 lwkt_reltoken(&mp->mnt_token); 196 197 return (error); 198 } 199 /* --------------------------------------------------------------------- */ 200 201 static int 202 tmpfs_nmknod(struct vop_nmknod_args *v) 203 { 204 struct vnode *dvp = v->a_dvp; 205 struct vnode **vpp = v->a_vpp; 206 struct namecache *ncp = v->a_nch->ncp; 207 struct vattr *vap = v->a_vap; 208 struct ucred *cred = v->a_cred; 209 struct mount *mp = dvp->v_mount; 210 int error; 211 212 lwkt_gettoken(&mp->mnt_token); 213 214 if (vap->va_type != VBLK && vap->va_type != VCHR && 215 vap->va_type != VFIFO) { 216 lwkt_reltoken(&mp->mnt_token); 217 return (EINVAL); 218 } 219 220 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 221 if (error == 0) { 222 cache_setunresolved(v->a_nch); 223 cache_setvp(v->a_nch, *vpp); 224 tmpfs_knote(dvp, NOTE_WRITE); 225 } 226 227 lwkt_reltoken(&mp->mnt_token); 228 229 return error; 230 } 231 232 /* --------------------------------------------------------------------- */ 233 234 static int 235 tmpfs_open(struct vop_open_args *v) 236 { 237 struct vnode *vp = v->a_vp; 238 int mode = v->a_mode; 239 struct mount *mp = vp->v_mount; 240 struct tmpfs_node *node; 241 int error; 242 243 lwkt_gettoken(&mp->mnt_token); 244 node = VP_TO_TMPFS_NODE(vp); 245 246 #if 0 247 /* The file is still active but all its names have been removed 248 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 249 * it is about to die. */ 250 if (node->tn_links < 1) 251 return (ENOENT); 252 #endif 253 254 /* If the file is marked append-only, deny write requests. */ 255 if ((node->tn_flags & APPEND) && 256 (mode & (FWRITE | O_APPEND)) == FWRITE) { 257 error = EPERM; 258 } else { 259 error = (vop_stdopen(v)); 260 } 261 262 lwkt_reltoken(&mp->mnt_token); 263 return (error); 264 } 265 266 /* --------------------------------------------------------------------- */ 267 268 static int 269 tmpfs_close(struct vop_close_args *v) 270 { 271 struct vnode *vp = v->a_vp; 272 struct tmpfs_node *node; 273 int error; 274 275 lwkt_gettoken(&vp->v_mount->mnt_token); 276 node = VP_TO_TMPFS_NODE(vp); 277 278 if (node->tn_links > 0) { 279 /* 280 * Update node times. No need to do it if the node has 281 * been deleted, because it will vanish after we return. 282 */ 283 tmpfs_update(vp); 284 } 285 286 error = vop_stdclose(v); 287 288 lwkt_reltoken(&vp->v_mount->mnt_token); 289 290 return (error); 291 } 292 293 /* --------------------------------------------------------------------- */ 294 295 int 296 tmpfs_access(struct vop_access_args *v) 297 { 298 struct vnode *vp = v->a_vp; 299 int error; 300 struct tmpfs_node *node; 301 302 lwkt_gettoken(&vp->v_mount->mnt_token); 303 node = VP_TO_TMPFS_NODE(vp); 304 305 switch (vp->v_type) { 306 case VDIR: 307 /* FALLTHROUGH */ 308 case VLNK: 309 /* FALLTHROUGH */ 310 case VREG: 311 if ((v->a_mode & VWRITE) && 312 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 313 error = EROFS; 314 goto out; 315 } 316 break; 317 318 case VBLK: 319 /* FALLTHROUGH */ 320 case VCHR: 321 /* FALLTHROUGH */ 322 case VSOCK: 323 /* FALLTHROUGH */ 324 case VFIFO: 325 break; 326 327 default: 328 error = EINVAL; 329 goto out; 330 } 331 332 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) { 333 error = EPERM; 334 goto out; 335 } 336 337 error = vop_helper_access(v, node->tn_uid, node->tn_gid, 338 node->tn_mode, 0); 339 340 out: 341 lwkt_reltoken(&vp->v_mount->mnt_token); 342 return error; 343 } 344 345 /* --------------------------------------------------------------------- */ 346 347 int 348 tmpfs_getattr(struct vop_getattr_args *v) 349 { 350 struct vnode *vp = v->a_vp; 351 struct vattr *vap = v->a_vap; 352 struct tmpfs_node *node; 353 354 lwkt_gettoken(&vp->v_mount->mnt_token); 355 node = VP_TO_TMPFS_NODE(vp); 356 357 tmpfs_update(vp); 358 359 vap->va_type = vp->v_type; 360 vap->va_mode = node->tn_mode; 361 vap->va_nlink = node->tn_links; 362 vap->va_uid = node->tn_uid; 363 vap->va_gid = node->tn_gid; 364 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 365 vap->va_fileid = node->tn_id; 366 vap->va_size = node->tn_size; 367 vap->va_blocksize = PAGE_SIZE; 368 vap->va_atime.tv_sec = node->tn_atime; 369 vap->va_atime.tv_nsec = node->tn_atimensec; 370 vap->va_mtime.tv_sec = node->tn_mtime; 371 vap->va_mtime.tv_nsec = node->tn_mtimensec; 372 vap->va_ctime.tv_sec = node->tn_ctime; 373 vap->va_ctime.tv_nsec = node->tn_ctimensec; 374 vap->va_gen = node->tn_gen; 375 vap->va_flags = node->tn_flags; 376 if (vp->v_type == VBLK || vp->v_type == VCHR) 377 { 378 vap->va_rmajor = umajor(node->tn_rdev); 379 vap->va_rminor = uminor(node->tn_rdev); 380 } 381 vap->va_bytes = round_page(node->tn_size); 382 vap->va_filerev = 0; 383 384 lwkt_reltoken(&vp->v_mount->mnt_token); 385 386 return 0; 387 } 388 389 /* --------------------------------------------------------------------- */ 390 391 int 392 tmpfs_setattr(struct vop_setattr_args *v) 393 { 394 struct vnode *vp = v->a_vp; 395 struct vattr *vap = v->a_vap; 396 struct ucred *cred = v->a_cred; 397 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp); 398 int error = 0; 399 int kflags = 0; 400 401 lwkt_gettoken(&vp->v_mount->mnt_token); 402 if (error == 0 && (vap->va_flags != VNOVAL)) { 403 error = tmpfs_chflags(vp, vap->va_flags, cred); 404 kflags |= NOTE_ATTRIB; 405 } 406 407 if (error == 0 && (vap->va_size != VNOVAL)) { 408 if (vap->va_size > node->tn_size) 409 kflags |= NOTE_WRITE | NOTE_EXTEND; 410 else 411 kflags |= NOTE_WRITE; 412 error = tmpfs_chsize(vp, vap->va_size, cred); 413 } 414 415 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL || 416 vap->va_gid != (gid_t)VNOVAL)) { 417 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred); 418 kflags |= NOTE_ATTRIB; 419 } 420 421 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) { 422 error = tmpfs_chmod(vp, vap->va_mode, cred); 423 kflags |= NOTE_ATTRIB; 424 } 425 426 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && 427 vap->va_atime.tv_nsec != VNOVAL) || 428 (vap->va_mtime.tv_sec != VNOVAL && 429 vap->va_mtime.tv_nsec != VNOVAL) )) { 430 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime, 431 vap->va_vaflags, cred); 432 kflags |= NOTE_ATTRIB; 433 } 434 435 /* Update the node times. We give preference to the error codes 436 * generated by this function rather than the ones that may arise 437 * from tmpfs_update. */ 438 tmpfs_update(vp); 439 tmpfs_knote(vp, kflags); 440 441 lwkt_reltoken(&vp->v_mount->mnt_token); 442 443 return (error); 444 } 445 446 /* --------------------------------------------------------------------- */ 447 448 /* 449 * fsync is usually a NOP, but we must take action when unmounting or 450 * when recycling. 451 */ 452 static int 453 tmpfs_fsync(struct vop_fsync_args *v) 454 { 455 struct tmpfs_node *node; 456 struct vnode *vp = v->a_vp; 457 458 lwkt_gettoken(&vp->v_mount->mnt_token); 459 node = VP_TO_TMPFS_NODE(vp); 460 461 tmpfs_update(vp); 462 if (vp->v_type == VREG) { 463 if (vp->v_flag & VRECLAIMED) { 464 if (node->tn_links == 0) 465 tmpfs_truncate(vp, 0); 466 else 467 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL); 468 } 469 } 470 471 lwkt_reltoken(&vp->v_mount->mnt_token); 472 return 0; 473 } 474 475 /* --------------------------------------------------------------------- */ 476 477 static int 478 tmpfs_read (struct vop_read_args *ap) 479 { 480 struct buf *bp; 481 struct vnode *vp = ap->a_vp; 482 struct uio *uio = ap->a_uio; 483 struct tmpfs_node *node; 484 off_t base_offset; 485 size_t offset; 486 size_t len; 487 size_t resid; 488 int error; 489 490 /* 491 * Check the basics 492 */ 493 if (uio->uio_offset < 0) 494 return (EINVAL); 495 if (vp->v_type != VREG) 496 return (EINVAL); 497 498 /* 499 * Extract node, try to shortcut the operation through 500 * the VM page cache, allowing us to avoid buffer cache 501 * overheads. 502 */ 503 node = VP_TO_TMPFS_NODE(vp); 504 resid = uio->uio_resid; 505 error = vop_helper_read_shortcut(ap); 506 if (error) 507 return error; 508 if (uio->uio_resid == 0) { 509 if (resid) 510 goto finished; 511 return error; 512 } 513 514 /* 515 * Fall-through to our normal read code. 516 */ 517 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) { 518 /* 519 * Use buffer cache I/O (via tmpfs_strategy) 520 */ 521 offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64; 522 base_offset = (off_t)uio->uio_offset - offset; 523 bp = getcacheblk(vp, base_offset, TMPFS_BLKSIZE, 0); 524 if (bp == NULL) { 525 lwkt_gettoken(&vp->v_mount->mnt_token); 526 error = bread(vp, base_offset, TMPFS_BLKSIZE, &bp); 527 if (error) { 528 brelse(bp); 529 lwkt_reltoken(&vp->v_mount->mnt_token); 530 kprintf("tmpfs_read bread error %d\n", error); 531 break; 532 } 533 lwkt_reltoken(&vp->v_mount->mnt_token); 534 535 /* 536 * tmpfs pretty much fiddles directly with the VM 537 * system, don't let it exhaust it or we won't play 538 * nice with other processes. 539 * 540 * Only do this if the VOP is coming from a normal 541 * read/write. The VM system handles the case for 542 * UIO_NOCOPY. 543 */ 544 if (uio->uio_segflg != UIO_NOCOPY) 545 vm_wait_nominal(); 546 } 547 bp->b_flags |= B_CLUSTEROK; 548 549 /* 550 * Figure out how many bytes we can actually copy this loop. 551 */ 552 len = TMPFS_BLKSIZE - offset; 553 if (len > uio->uio_resid) 554 len = uio->uio_resid; 555 if (len > node->tn_size - uio->uio_offset) 556 len = (size_t)(node->tn_size - uio->uio_offset); 557 558 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio); 559 bqrelse(bp); 560 if (error) { 561 kprintf("tmpfs_read uiomove error %d\n", error); 562 break; 563 } 564 } 565 566 finished: 567 TMPFS_NODE_LOCK(node); 568 node->tn_status |= TMPFS_NODE_ACCESSED; 569 TMPFS_NODE_UNLOCK(node); 570 571 return (error); 572 } 573 574 static int 575 tmpfs_write (struct vop_write_args *ap) 576 { 577 struct buf *bp; 578 struct vnode *vp = ap->a_vp; 579 struct uio *uio = ap->a_uio; 580 struct thread *td = uio->uio_td; 581 struct tmpfs_node *node; 582 boolean_t extended; 583 off_t oldsize; 584 int error; 585 off_t base_offset; 586 size_t offset; 587 size_t len; 588 struct rlimit limit; 589 int trivial = 0; 590 int kflags = 0; 591 int seqcount; 592 593 error = 0; 594 if (uio->uio_resid == 0) { 595 return error; 596 } 597 598 node = VP_TO_TMPFS_NODE(vp); 599 600 if (vp->v_type != VREG) 601 return (EINVAL); 602 seqcount = ap->a_ioflag >> 16; 603 604 lwkt_gettoken(&vp->v_mount->mnt_token); 605 606 oldsize = node->tn_size; 607 if (ap->a_ioflag & IO_APPEND) 608 uio->uio_offset = node->tn_size; 609 610 /* 611 * Check for illegal write offsets. 612 */ 613 if (uio->uio_offset + uio->uio_resid > 614 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) { 615 lwkt_reltoken(&vp->v_mount->mnt_token); 616 return (EFBIG); 617 } 618 619 /* 620 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN). 621 */ 622 if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) { 623 error = kern_getrlimit(RLIMIT_FSIZE, &limit); 624 if (error != 0) { 625 lwkt_reltoken(&vp->v_mount->mnt_token); 626 return error; 627 } 628 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) { 629 ksignal(td->td_proc, SIGXFSZ); 630 lwkt_reltoken(&vp->v_mount->mnt_token); 631 return (EFBIG); 632 } 633 } 634 635 636 /* 637 * Extend the file's size if necessary 638 */ 639 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size); 640 641 while (uio->uio_resid > 0) { 642 /* 643 * Don't completely blow out running buffer I/O 644 * when being hit from the pageout daemon. 645 */ 646 if (uio->uio_segflg == UIO_NOCOPY && 647 (ap->a_ioflag & IO_RECURSE) == 0) { 648 bwillwrite(TMPFS_BLKSIZE); 649 } 650 651 /* 652 * Use buffer cache I/O (via tmpfs_strategy) 653 */ 654 offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64; 655 base_offset = (off_t)uio->uio_offset - offset; 656 len = TMPFS_BLKSIZE - offset; 657 if (len > uio->uio_resid) 658 len = uio->uio_resid; 659 660 if ((uio->uio_offset + len) > node->tn_size) { 661 trivial = (uio->uio_offset <= node->tn_size); 662 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial); 663 if (error) 664 break; 665 } 666 667 /* 668 * Read to fill in any gaps. Theoretically we could 669 * optimize this if the write covers the entire buffer 670 * and is not a UIO_NOCOPY write, however this can lead 671 * to a security violation exposing random kernel memory 672 * (whatever junk was in the backing VM pages before). 673 * 674 * So just use bread() to do the right thing. 675 */ 676 error = bread(vp, base_offset, TMPFS_BLKSIZE, &bp); 677 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio); 678 if (error) { 679 kprintf("tmpfs_write uiomove error %d\n", error); 680 brelse(bp); 681 break; 682 } 683 684 if (uio->uio_offset > node->tn_size) { 685 node->tn_size = uio->uio_offset; 686 kflags |= NOTE_EXTEND; 687 } 688 kflags |= NOTE_WRITE; 689 690 /* 691 * Always try to flush the page in the UIO_NOCOPY case. This 692 * can come from the pageout daemon or during vnode eviction. 693 * It is not necessarily going to be marked IO_ASYNC/IO_SYNC. 694 * 695 * For the normal case we buwrite(), dirtying the underlying 696 * VM pages instead of dirtying the buffer and releasing the 697 * buffer as a clean buffer. This allows tmpfs to use 698 * essentially all available memory to cache file data. 699 * If we used bdwrite() the buffer cache would wind up 700 * flushing the data to swap too quickly. 701 * 702 * But because tmpfs can seriously load the VM system we 703 * fall-back to using bdwrite() when free memory starts 704 * to get low. This shifts the load away from the VM system 705 * and makes tmpfs act more like a normal filesystem with 706 * regards to disk activity. 707 * 708 * tmpfs pretty much fiddles directly with the VM 709 * system, don't let it exhaust it or we won't play 710 * nice with other processes. Only do this if the 711 * VOP is coming from a normal read/write. The VM system 712 * handles the case for UIO_NOCOPY. 713 */ 714 bp->b_flags |= B_CLUSTEROK; 715 if (uio->uio_segflg == UIO_NOCOPY) { 716 /* 717 * Flush from the pageout daemon, deal with 718 * potentially very heavy tmpfs write activity 719 * causing long stalls in the pageout daemon 720 * before pages get to free/cache. 721 * 722 * (a) Under severe pressure setting B_DIRECT will 723 * cause a buffer release to try to free the 724 * underlying pages. 725 * 726 * (b) Under modest memory pressure the B_RELBUF 727 * alone is sufficient to get the pages moved 728 * to the cache. We could also force this by 729 * setting B_NOTMETA but that might have other 730 * unintended side-effects (e.g. setting 731 * PG_NOTMETA on the VM page). 732 * 733 * Hopefully this will unblock the VM system more 734 * quickly under extreme tmpfs write load. 735 */ 736 if (vm_page_count_min(vm_page_free_hysteresis)) 737 bp->b_flags |= B_DIRECT; 738 bp->b_flags |= B_AGE | B_RELBUF; 739 bp->b_act_count = 0; /* buffer->deactivate pgs */ 740 cluster_awrite(bp); 741 } else if (vm_page_count_target()) { 742 /* 743 * Normal (userland) write but we are low on memory, 744 * run the buffer the buffer cache. 745 */ 746 bp->b_act_count = 0; /* buffer->deactivate pgs */ 747 bdwrite(bp); 748 } else { 749 /* 750 * Otherwise run the buffer directly through to the 751 * backing VM store. 752 */ 753 buwrite(bp); 754 /*vm_wait_nominal();*/ 755 } 756 757 if (bp->b_error) { 758 kprintf("tmpfs_write bwrite error %d\n", bp->b_error); 759 break; 760 } 761 } 762 763 if (error) { 764 if (extended) { 765 (void)tmpfs_reg_resize(vp, oldsize, trivial); 766 kflags &= ~NOTE_EXTEND; 767 } 768 goto done; 769 } 770 771 /* 772 * Currently we don't set the mtime on files modified via mmap() 773 * because we can't tell the difference between those modifications 774 * and an attempt by the pageout daemon to flush tmpfs pages to 775 * swap. 776 * 777 * This is because in order to defer flushes as long as possible 778 * buwrite() works by marking the underlying VM pages dirty in 779 * order to be able to dispose of the buffer cache buffer without 780 * flushing it. 781 */ 782 TMPFS_NODE_LOCK(node); 783 if (uio->uio_segflg != UIO_NOCOPY) 784 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED; 785 if (extended) 786 node->tn_status |= TMPFS_NODE_CHANGED; 787 788 if (node->tn_mode & (S_ISUID | S_ISGID)) { 789 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) 790 node->tn_mode &= ~(S_ISUID | S_ISGID); 791 } 792 TMPFS_NODE_UNLOCK(node); 793 done: 794 795 tmpfs_knote(vp, kflags); 796 797 lwkt_reltoken(&vp->v_mount->mnt_token); 798 return(error); 799 } 800 801 static int 802 tmpfs_advlock (struct vop_advlock_args *ap) 803 { 804 struct tmpfs_node *node; 805 struct vnode *vp = ap->a_vp; 806 int error; 807 808 lwkt_gettoken(&vp->v_mount->mnt_token); 809 node = VP_TO_TMPFS_NODE(vp); 810 811 error = (lf_advlock(ap, &node->tn_advlock, node->tn_size)); 812 lwkt_reltoken(&vp->v_mount->mnt_token); 813 814 return (error); 815 } 816 817 /* 818 * The strategy function is typically only called when memory pressure 819 * forces the system to attempt to pageout pages. It can also be called 820 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write 821 * operations 822 */ 823 static int 824 tmpfs_strategy(struct vop_strategy_args *ap) 825 { 826 struct bio *bio = ap->a_bio; 827 struct bio *nbio; 828 struct buf *bp = bio->bio_buf; 829 struct vnode *vp = ap->a_vp; 830 struct tmpfs_node *node; 831 vm_object_t uobj; 832 vm_page_t m; 833 int i; 834 835 if (vp->v_type != VREG) { 836 bp->b_resid = bp->b_bcount; 837 bp->b_flags |= B_ERROR | B_INVAL; 838 bp->b_error = EINVAL; 839 biodone(bio); 840 return(0); 841 } 842 843 lwkt_gettoken(&vp->v_mount->mnt_token); 844 node = VP_TO_TMPFS_NODE(vp); 845 846 uobj = node->tn_reg.tn_aobj; 847 848 /* 849 * Don't bother flushing to swap if there is no swap, just 850 * ensure that the pages are marked as needing a commit (still). 851 */ 852 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) { 853 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 854 m = bp->b_xio.xio_pages[i]; 855 vm_page_need_commit(m); 856 } 857 bp->b_resid = 0; 858 bp->b_error = 0; 859 biodone(bio); 860 } else { 861 nbio = push_bio(bio); 862 nbio->bio_done = tmpfs_strategy_done; 863 nbio->bio_offset = bio->bio_offset; 864 swap_pager_strategy(uobj, nbio); 865 } 866 867 lwkt_reltoken(&vp->v_mount->mnt_token); 868 return 0; 869 } 870 871 /* 872 * If we were unable to commit the pages to swap make sure they are marked 873 * as needing a commit (again). If we were, clear the flag to allow the 874 * pages to be freed. 875 */ 876 static void 877 tmpfs_strategy_done(struct bio *bio) 878 { 879 struct buf *bp; 880 vm_page_t m; 881 int i; 882 883 bp = bio->bio_buf; 884 885 if (bp->b_flags & B_ERROR) { 886 bp->b_flags &= ~B_ERROR; 887 bp->b_error = 0; 888 bp->b_resid = 0; 889 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 890 m = bp->b_xio.xio_pages[i]; 891 vm_page_need_commit(m); 892 } 893 } else { 894 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 895 m = bp->b_xio.xio_pages[i]; 896 vm_page_clear_commit(m); 897 } 898 } 899 bio = pop_bio(bio); 900 biodone(bio); 901 } 902 903 static int 904 tmpfs_bmap(struct vop_bmap_args *ap) 905 { 906 if (ap->a_doffsetp != NULL) 907 *ap->a_doffsetp = ap->a_loffset; 908 if (ap->a_runp != NULL) 909 *ap->a_runp = 0; 910 if (ap->a_runb != NULL) 911 *ap->a_runb = 0; 912 913 return 0; 914 } 915 916 /* --------------------------------------------------------------------- */ 917 918 static int 919 tmpfs_nremove(struct vop_nremove_args *v) 920 { 921 struct vnode *dvp = v->a_dvp; 922 struct namecache *ncp = v->a_nch->ncp; 923 struct vnode *vp; 924 int error; 925 struct tmpfs_dirent *de; 926 struct tmpfs_mount *tmp; 927 struct tmpfs_node *dnode; 928 struct tmpfs_node *node; 929 struct mount *mp; 930 931 mp = dvp->v_mount; 932 933 lwkt_gettoken(&mp->mnt_token); 934 935 /* 936 * We have to acquire the vp from v->a_nch because we will likely 937 * unresolve the namecache entry, and a vrele/vput is needed to 938 * trigger the tmpfs_inactive/tmpfs_reclaim sequence. 939 * 940 * We have to use vget to clear any inactive state on the vnode, 941 * otherwise the vnode may remain inactive and thus tmpfs_inactive 942 * will not get called when we release it. 943 */ 944 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp); 945 KKASSERT(vp->v_mount == dvp->v_mount); 946 KKASSERT(error == 0); 947 vn_unlock(vp); 948 949 if (vp->v_type == VDIR) { 950 error = EISDIR; 951 goto out; 952 } 953 954 dnode = VP_TO_TMPFS_DIR(dvp); 955 node = VP_TO_TMPFS_NODE(vp); 956 tmp = VFS_TO_TMPFS(vp->v_mount); 957 de = tmpfs_dir_lookup(dnode, node, ncp); 958 if (de == NULL) { 959 error = ENOENT; 960 goto out; 961 } 962 963 /* Files marked as immutable or append-only cannot be deleted. */ 964 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 965 (dnode->tn_flags & APPEND)) { 966 error = EPERM; 967 goto out; 968 } 969 970 /* Remove the entry from the directory; as it is a file, we do not 971 * have to change the number of hard links of the directory. */ 972 tmpfs_dir_detach(dnode, de); 973 974 /* Free the directory entry we just deleted. Note that the node 975 * referred by it will not be removed until the vnode is really 976 * reclaimed. */ 977 tmpfs_free_dirent(tmp, de); 978 979 if (node->tn_links > 0) { 980 TMPFS_NODE_LOCK(node); 981 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 982 TMPFS_NODE_MODIFIED; 983 TMPFS_NODE_UNLOCK(node); 984 } 985 986 cache_unlink(v->a_nch); 987 tmpfs_knote(vp, NOTE_DELETE); 988 tmpfs_knote(dvp, NOTE_WRITE); 989 error = 0; 990 991 out: 992 vrele(vp); 993 lwkt_reltoken(&mp->mnt_token); 994 995 return error; 996 } 997 998 /* --------------------------------------------------------------------- */ 999 1000 static int 1001 tmpfs_nlink(struct vop_nlink_args *v) 1002 { 1003 struct vnode *dvp = v->a_dvp; 1004 struct vnode *vp = v->a_vp; 1005 struct namecache *ncp = v->a_nch->ncp; 1006 struct tmpfs_dirent *de; 1007 struct tmpfs_node *node; 1008 struct tmpfs_node *dnode; 1009 struct mount *mp; 1010 int error; 1011 1012 if (dvp->v_mount != vp->v_mount) 1013 return(EXDEV); 1014 mp = dvp->v_mount; 1015 1016 lwkt_gettoken(&mp->mnt_token); 1017 KKASSERT(dvp != vp); /* XXX When can this be false? */ 1018 1019 node = VP_TO_TMPFS_NODE(vp); 1020 dnode = VP_TO_TMPFS_NODE(dvp); 1021 1022 /* XXX: Why aren't the following two tests done by the caller? */ 1023 1024 /* Hard links of directories are forbidden. */ 1025 if (vp->v_type == VDIR) { 1026 error = EPERM; 1027 goto out; 1028 } 1029 1030 /* Cannot create cross-device links. */ 1031 if (dvp->v_mount != vp->v_mount) { 1032 error = EXDEV; 1033 goto out; 1034 } 1035 1036 /* Ensure that we do not overflow the maximum number of links imposed 1037 * by the system. */ 1038 KKASSERT(node->tn_links <= LINK_MAX); 1039 if (node->tn_links == LINK_MAX) { 1040 error = EMLINK; 1041 goto out; 1042 } 1043 1044 /* We cannot create links of files marked immutable or append-only. */ 1045 if (node->tn_flags & (IMMUTABLE | APPEND)) { 1046 error = EPERM; 1047 goto out; 1048 } 1049 1050 /* Allocate a new directory entry to represent the node. */ 1051 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 1052 ncp->nc_name, ncp->nc_nlen, &de); 1053 if (error != 0) 1054 goto out; 1055 1056 /* Insert the new directory entry into the appropriate directory. */ 1057 tmpfs_dir_attach(dnode, de); 1058 1059 /* vp link count has changed, so update node times. */ 1060 1061 TMPFS_NODE_LOCK(node); 1062 node->tn_status |= TMPFS_NODE_CHANGED; 1063 TMPFS_NODE_UNLOCK(node); 1064 tmpfs_update(vp); 1065 1066 tmpfs_knote(vp, NOTE_LINK); 1067 cache_setunresolved(v->a_nch); 1068 cache_setvp(v->a_nch, vp); 1069 tmpfs_knote(dvp, NOTE_WRITE); 1070 error = 0; 1071 1072 out: 1073 lwkt_reltoken(&mp->mnt_token); 1074 return error; 1075 } 1076 1077 /* --------------------------------------------------------------------- */ 1078 1079 static int 1080 tmpfs_nrename(struct vop_nrename_args *v) 1081 { 1082 struct vnode *fdvp = v->a_fdvp; 1083 struct namecache *fncp = v->a_fnch->ncp; 1084 struct vnode *fvp = fncp->nc_vp; 1085 struct vnode *tdvp = v->a_tdvp; 1086 struct namecache *tncp = v->a_tnch->ncp; 1087 struct vnode *tvp; 1088 struct tmpfs_dirent *de, *tde; 1089 struct tmpfs_mount *tmp; 1090 struct tmpfs_node *fdnode; 1091 struct tmpfs_node *fnode; 1092 struct tmpfs_node *tnode; 1093 struct tmpfs_node *tdnode; 1094 struct mount *mp; 1095 char *newname; 1096 char *oldname; 1097 int error; 1098 1099 mp = fdvp->v_mount; 1100 KKASSERT(fdvp->v_mount == fvp->v_mount); 1101 1102 lwkt_gettoken(&mp->mnt_token); 1103 /* 1104 * Because tvp can get overwritten we have to vget it instead of 1105 * just vref or use it, otherwise it's VINACTIVE flag may not get 1106 * cleared and the node won't get destroyed. 1107 */ 1108 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp); 1109 if (error == 0) { 1110 tnode = VP_TO_TMPFS_NODE(tvp); 1111 vn_unlock(tvp); 1112 } else { 1113 tnode = NULL; 1114 } 1115 1116 /* Disallow cross-device renames. 1117 * XXX Why isn't this done by the caller? */ 1118 if (fvp->v_mount != tdvp->v_mount || 1119 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 1120 error = EXDEV; 1121 goto out; 1122 } 1123 1124 tmp = VFS_TO_TMPFS(tdvp->v_mount); 1125 tdnode = VP_TO_TMPFS_DIR(tdvp); 1126 1127 /* If source and target are the same file, there is nothing to do. */ 1128 if (fvp == tvp) { 1129 error = 0; 1130 goto out; 1131 } 1132 1133 fdnode = VP_TO_TMPFS_DIR(fdvp); 1134 fnode = VP_TO_TMPFS_NODE(fvp); 1135 de = tmpfs_dir_lookup(fdnode, fnode, fncp); 1136 1137 /* Avoid manipulating '.' and '..' entries. */ 1138 if (de == NULL) { 1139 error = ENOENT; 1140 goto out_locked; 1141 } 1142 KKASSERT(de->td_node == fnode); 1143 1144 /* 1145 * If replacing an entry in the target directory and that entry 1146 * is a directory, it must be empty. 1147 * 1148 * Kern_rename gurantees the destination to be a directory 1149 * if the source is one (it does?). 1150 */ 1151 if (tvp != NULL) { 1152 KKASSERT(tnode != NULL); 1153 1154 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1155 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 1156 error = EPERM; 1157 goto out_locked; 1158 } 1159 1160 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { 1161 if (tnode->tn_size > 0) { 1162 error = ENOTEMPTY; 1163 goto out_locked; 1164 } 1165 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { 1166 error = ENOTDIR; 1167 goto out_locked; 1168 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { 1169 error = EISDIR; 1170 goto out_locked; 1171 } else { 1172 KKASSERT(fnode->tn_type != VDIR && 1173 tnode->tn_type != VDIR); 1174 } 1175 } 1176 1177 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1178 (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 1179 error = EPERM; 1180 goto out_locked; 1181 } 1182 1183 /* 1184 * Ensure that we have enough memory to hold the new name, if it 1185 * has to be changed. 1186 */ 1187 if (fncp->nc_nlen != tncp->nc_nlen || 1188 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) { 1189 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone, 1190 M_WAITOK | M_NULLOK); 1191 if (newname == NULL) { 1192 error = ENOSPC; 1193 goto out_locked; 1194 } 1195 bcopy(tncp->nc_name, newname, tncp->nc_nlen); 1196 newname[tncp->nc_nlen] = '\0'; 1197 } else { 1198 newname = NULL; 1199 } 1200 1201 /* 1202 * Unlink entry from source directory. Note that the kernel has 1203 * already checked for illegal recursion cases (renaming a directory 1204 * into a subdirectory of itself). 1205 */ 1206 if (fdnode != tdnode) { 1207 tmpfs_dir_detach(fdnode, de); 1208 } else { 1209 RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de); 1210 } 1211 1212 /* 1213 * Handle any name change. Swap with newname, we will 1214 * deallocate it at the end. 1215 */ 1216 if (newname != NULL) { 1217 #if 0 1218 TMPFS_NODE_LOCK(fnode); 1219 fnode->tn_status |= TMPFS_NODE_CHANGED; 1220 TMPFS_NODE_UNLOCK(fnode); 1221 #endif 1222 oldname = de->td_name; 1223 de->td_name = newname; 1224 de->td_namelen = (uint16_t)tncp->nc_nlen; 1225 newname = oldname; 1226 } 1227 1228 /* 1229 * If we are overwriting an entry, we have to remove the old one 1230 * from the target directory. 1231 */ 1232 if (tvp != NULL) { 1233 /* Remove the old entry from the target directory. */ 1234 tde = tmpfs_dir_lookup(tdnode, tnode, tncp); 1235 tmpfs_dir_detach(tdnode, tde); 1236 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE); 1237 1238 /* 1239 * Free the directory entry we just deleted. Note that the 1240 * node referred by it will not be removed until the vnode is 1241 * really reclaimed. 1242 */ 1243 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde); 1244 /*cache_inval_vp(tvp, CINV_DESTROY);*/ 1245 } 1246 1247 /* 1248 * Link entry to target directory. If the entry 1249 * represents a directory move the parent linkage 1250 * as well. 1251 */ 1252 if (fdnode != tdnode) { 1253 if (de->td_node->tn_type == VDIR) { 1254 TMPFS_VALIDATE_DIR(fnode); 1255 } 1256 tmpfs_dir_attach(tdnode, de); 1257 } else { 1258 TMPFS_NODE_LOCK(tdnode); 1259 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 1260 RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de); 1261 TMPFS_NODE_UNLOCK(tdnode); 1262 } 1263 1264 /* 1265 * Finish up 1266 */ 1267 if (newname) { 1268 kfree(newname, tmp->tm_name_zone); 1269 newname = NULL; 1270 } 1271 cache_rename(v->a_fnch, v->a_tnch); 1272 tmpfs_knote(v->a_fdvp, NOTE_WRITE); 1273 tmpfs_knote(v->a_tdvp, NOTE_WRITE); 1274 if (fnode->tn_vnode) 1275 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME); 1276 error = 0; 1277 1278 out_locked: 1279 ; 1280 1281 out: 1282 if (tvp) 1283 vrele(tvp); 1284 1285 lwkt_reltoken(&mp->mnt_token); 1286 1287 return error; 1288 } 1289 1290 /* --------------------------------------------------------------------- */ 1291 1292 static int 1293 tmpfs_nmkdir(struct vop_nmkdir_args *v) 1294 { 1295 struct vnode *dvp = v->a_dvp; 1296 struct vnode **vpp = v->a_vpp; 1297 struct namecache *ncp = v->a_nch->ncp; 1298 struct vattr *vap = v->a_vap; 1299 struct ucred *cred = v->a_cred; 1300 struct mount *mp; 1301 int error; 1302 1303 mp = dvp->v_mount; 1304 1305 lwkt_gettoken(&mp->mnt_token); 1306 KKASSERT(vap->va_type == VDIR); 1307 1308 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 1309 if (error == 0) { 1310 cache_setunresolved(v->a_nch); 1311 cache_setvp(v->a_nch, *vpp); 1312 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 1313 } 1314 1315 lwkt_reltoken(&mp->mnt_token); 1316 1317 return error; 1318 } 1319 1320 /* --------------------------------------------------------------------- */ 1321 1322 static int 1323 tmpfs_nrmdir(struct vop_nrmdir_args *v) 1324 { 1325 struct vnode *dvp = v->a_dvp; 1326 struct namecache *ncp = v->a_nch->ncp; 1327 struct vnode *vp; 1328 struct tmpfs_dirent *de; 1329 struct tmpfs_mount *tmp; 1330 struct tmpfs_node *dnode; 1331 struct tmpfs_node *node; 1332 struct mount *mp; 1333 int error; 1334 1335 mp = dvp->v_mount; 1336 lwkt_gettoken(&mp->mnt_token); 1337 1338 /* 1339 * We have to acquire the vp from v->a_nch because we will likely 1340 * unresolve the namecache entry, and a vrele/vput is needed to 1341 * trigger the tmpfs_inactive/tmpfs_reclaim sequence. 1342 * 1343 * We have to use vget to clear any inactive state on the vnode, 1344 * otherwise the vnode may remain inactive and thus tmpfs_inactive 1345 * will not get called when we release it. 1346 */ 1347 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp); 1348 KKASSERT(error == 0); 1349 vn_unlock(vp); 1350 1351 /* 1352 * Prevalidate so we don't hit an assertion later 1353 */ 1354 if (vp->v_type != VDIR) { 1355 error = ENOTDIR; 1356 goto out; 1357 } 1358 1359 tmp = VFS_TO_TMPFS(dvp->v_mount); 1360 dnode = VP_TO_TMPFS_DIR(dvp); 1361 node = VP_TO_TMPFS_DIR(vp); 1362 1363 /* Directories with more than two entries ('.' and '..') cannot be 1364 * removed. */ 1365 if (node->tn_size > 0) { 1366 error = ENOTEMPTY; 1367 goto out; 1368 } 1369 1370 if ((dnode->tn_flags & APPEND) 1371 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1372 error = EPERM; 1373 goto out; 1374 } 1375 1376 /* This invariant holds only if we are not trying to remove "..". 1377 * We checked for that above so this is safe now. */ 1378 KKASSERT(node->tn_dir.tn_parent == dnode); 1379 1380 /* Get the directory entry associated with node (vp). This was 1381 * filled by tmpfs_lookup while looking up the entry. */ 1382 de = tmpfs_dir_lookup(dnode, node, ncp); 1383 KKASSERT(TMPFS_DIRENT_MATCHES(de, 1384 ncp->nc_name, 1385 ncp->nc_nlen)); 1386 1387 /* Check flags to see if we are allowed to remove the directory. */ 1388 if ((dnode->tn_flags & APPEND) || 1389 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { 1390 error = EPERM; 1391 goto out; 1392 } 1393 1394 1395 /* Detach the directory entry from the directory (dnode). */ 1396 tmpfs_dir_detach(dnode, de); 1397 1398 /* No vnode should be allocated for this entry from this point */ 1399 TMPFS_NODE_LOCK(node); 1400 TMPFS_ASSERT_ELOCKED(node); 1401 TMPFS_NODE_LOCK(dnode); 1402 TMPFS_ASSERT_ELOCKED(dnode); 1403 1404 /* 1405 * Must set parent linkage to NULL (tested by ncreate to disallow 1406 * the creation of new files/dirs in a deleted directory) 1407 */ 1408 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1409 TMPFS_NODE_MODIFIED; 1410 1411 dnode->tn_status |= TMPFS_NODE_ACCESSED | \ 1412 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1413 1414 TMPFS_NODE_UNLOCK(dnode); 1415 TMPFS_NODE_UNLOCK(node); 1416 1417 /* Free the directory entry we just deleted. Note that the node 1418 * referred by it will not be removed until the vnode is really 1419 * reclaimed. */ 1420 tmpfs_free_dirent(tmp, de); 1421 1422 /* Release the deleted vnode (will destroy the node, notify 1423 * interested parties and clean it from the cache). */ 1424 1425 TMPFS_NODE_LOCK(dnode); 1426 dnode->tn_status |= TMPFS_NODE_CHANGED; 1427 TMPFS_NODE_UNLOCK(dnode); 1428 tmpfs_update(dvp); 1429 1430 cache_unlink(v->a_nch); 1431 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 1432 error = 0; 1433 1434 out: 1435 vrele(vp); 1436 1437 lwkt_reltoken(&mp->mnt_token); 1438 1439 return error; 1440 } 1441 1442 /* --------------------------------------------------------------------- */ 1443 1444 static int 1445 tmpfs_nsymlink(struct vop_nsymlink_args *v) 1446 { 1447 struct vnode *dvp = v->a_dvp; 1448 struct vnode **vpp = v->a_vpp; 1449 struct namecache *ncp = v->a_nch->ncp; 1450 struct vattr *vap = v->a_vap; 1451 struct ucred *cred = v->a_cred; 1452 char *target = v->a_target; 1453 struct mount *mp = dvp->v_mount; 1454 int error; 1455 1456 lwkt_gettoken(&mp->mnt_token); 1457 vap->va_type = VLNK; 1458 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target); 1459 if (error == 0) { 1460 tmpfs_knote(*vpp, NOTE_WRITE); 1461 cache_setunresolved(v->a_nch); 1462 cache_setvp(v->a_nch, *vpp); 1463 } 1464 1465 lwkt_reltoken(&mp->mnt_token); 1466 1467 return error; 1468 } 1469 1470 /* --------------------------------------------------------------------- */ 1471 1472 static int 1473 tmpfs_readdir(struct vop_readdir_args *v) 1474 { 1475 struct vnode *vp = v->a_vp; 1476 struct uio *uio = v->a_uio; 1477 int *eofflag = v->a_eofflag; 1478 off_t **cookies = v->a_cookies; 1479 int *ncookies = v->a_ncookies; 1480 struct tmpfs_mount *tmp; 1481 int error; 1482 off_t startoff; 1483 off_t cnt = 0; 1484 struct tmpfs_node *node; 1485 struct mount *mp = vp->v_mount; 1486 1487 lwkt_gettoken(&mp->mnt_token); 1488 1489 /* This operation only makes sense on directory nodes. */ 1490 if (vp->v_type != VDIR) { 1491 lwkt_reltoken(&mp->mnt_token); 1492 return ENOTDIR; 1493 } 1494 1495 tmp = VFS_TO_TMPFS(vp->v_mount); 1496 node = VP_TO_TMPFS_DIR(vp); 1497 startoff = uio->uio_offset; 1498 1499 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) { 1500 error = tmpfs_dir_getdotdent(node, uio); 1501 if (error != 0) 1502 goto outok; 1503 cnt++; 1504 } 1505 1506 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) { 1507 error = tmpfs_dir_getdotdotdent(tmp, node, uio); 1508 if (error != 0) 1509 goto outok; 1510 cnt++; 1511 } 1512 1513 error = tmpfs_dir_getdents(node, uio, &cnt); 1514 1515 outok: 1516 KKASSERT(error >= -1); 1517 1518 if (error == -1) 1519 error = 0; 1520 1521 if (eofflag != NULL) 1522 *eofflag = 1523 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1524 1525 /* Update NFS-related variables. */ 1526 if (error == 0 && cookies != NULL && ncookies != NULL) { 1527 off_t i; 1528 off_t off = startoff; 1529 struct tmpfs_dirent *de = NULL; 1530 1531 *ncookies = cnt; 1532 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK); 1533 1534 for (i = 0; i < cnt; i++) { 1535 KKASSERT(off != TMPFS_DIRCOOKIE_EOF); 1536 if (off == TMPFS_DIRCOOKIE_DOT) { 1537 off = TMPFS_DIRCOOKIE_DOTDOT; 1538 } else { 1539 if (off == TMPFS_DIRCOOKIE_DOTDOT) { 1540 de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree); 1541 } else if (de != NULL) { 1542 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de); 1543 } else { 1544 de = tmpfs_dir_lookupbycookie(node, 1545 off); 1546 KKASSERT(de != NULL); 1547 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de); 1548 } 1549 if (de == NULL) 1550 off = TMPFS_DIRCOOKIE_EOF; 1551 else 1552 off = tmpfs_dircookie(de); 1553 } 1554 1555 (*cookies)[i] = off; 1556 } 1557 KKASSERT(uio->uio_offset == off); 1558 } 1559 1560 lwkt_reltoken(&mp->mnt_token); 1561 1562 return error; 1563 } 1564 1565 /* --------------------------------------------------------------------- */ 1566 1567 static int 1568 tmpfs_readlink(struct vop_readlink_args *v) 1569 { 1570 struct vnode *vp = v->a_vp; 1571 struct uio *uio = v->a_uio; 1572 struct mount *mp = vp->v_mount; 1573 int error; 1574 struct tmpfs_node *node; 1575 1576 lwkt_gettoken(&mp->mnt_token); 1577 1578 KKASSERT(uio->uio_offset == 0); 1579 KKASSERT(vp->v_type == VLNK); 1580 1581 node = VP_TO_TMPFS_NODE(vp); 1582 1583 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid), 1584 uio); 1585 TMPFS_NODE_LOCK(node); 1586 node->tn_status |= TMPFS_NODE_ACCESSED; 1587 TMPFS_NODE_UNLOCK(node); 1588 1589 lwkt_reltoken(&mp->mnt_token); 1590 1591 return error; 1592 } 1593 1594 /* --------------------------------------------------------------------- */ 1595 1596 static int 1597 tmpfs_inactive(struct vop_inactive_args *v) 1598 { 1599 struct vnode *vp = v->a_vp; 1600 struct tmpfs_node *node; 1601 struct mount *mp; 1602 1603 mp = vp->v_mount; 1604 lwkt_gettoken(&mp->mnt_token); 1605 node = VP_TO_TMPFS_NODE(vp); 1606 1607 /* 1608 * Degenerate case 1609 */ 1610 if (node == NULL) { 1611 vrecycle(vp); 1612 lwkt_reltoken(&mp->mnt_token); 1613 return(0); 1614 } 1615 1616 /* 1617 * Get rid of unreferenced deleted vnodes sooner rather than 1618 * later so the data memory can be recovered immediately. 1619 * 1620 * We must truncate the vnode to prevent the normal reclamation 1621 * path from flushing the data for the removed file to disk. 1622 */ 1623 TMPFS_NODE_LOCK(node); 1624 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && 1625 node->tn_links == 0) 1626 { 1627 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1628 TMPFS_NODE_UNLOCK(node); 1629 if (node->tn_type == VREG) 1630 tmpfs_truncate(vp, 0); 1631 vrecycle(vp); 1632 } else { 1633 TMPFS_NODE_UNLOCK(node); 1634 } 1635 lwkt_reltoken(&mp->mnt_token); 1636 1637 return 0; 1638 } 1639 1640 /* --------------------------------------------------------------------- */ 1641 1642 int 1643 tmpfs_reclaim(struct vop_reclaim_args *v) 1644 { 1645 struct vnode *vp = v->a_vp; 1646 struct tmpfs_mount *tmp; 1647 struct tmpfs_node *node; 1648 struct mount *mp; 1649 1650 mp = vp->v_mount; 1651 lwkt_gettoken(&mp->mnt_token); 1652 1653 node = VP_TO_TMPFS_NODE(vp); 1654 tmp = VFS_TO_TMPFS(vp->v_mount); 1655 KKASSERT(mp == tmp->tm_mount); 1656 1657 tmpfs_free_vp(vp); 1658 1659 /* 1660 * If the node referenced by this vnode was deleted by the 1661 * user, we must free its associated data structures now that 1662 * the vnode is being reclaimed. 1663 * 1664 * Directories have an extra link ref. 1665 */ 1666 TMPFS_NODE_LOCK(node); 1667 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && 1668 node->tn_links == 0) { 1669 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1670 tmpfs_free_node(tmp, node); 1671 /* eats the lock */ 1672 } else { 1673 TMPFS_NODE_UNLOCK(node); 1674 } 1675 lwkt_reltoken(&mp->mnt_token); 1676 1677 KKASSERT(vp->v_data == NULL); 1678 return 0; 1679 } 1680 1681 /* --------------------------------------------------------------------- */ 1682 1683 static int 1684 tmpfs_mountctl(struct vop_mountctl_args *ap) 1685 { 1686 struct tmpfs_mount *tmp; 1687 struct mount *mp; 1688 int rc; 1689 1690 mp = ap->a_head.a_ops->head.vv_mount; 1691 lwkt_gettoken(&mp->mnt_token); 1692 1693 switch (ap->a_op) { 1694 case (MOUNTCTL_SET_EXPORT): 1695 tmp = (struct tmpfs_mount *) mp->mnt_data; 1696 1697 if (ap->a_ctllen != sizeof(struct export_args)) 1698 rc = (EINVAL); 1699 else 1700 rc = vfs_export(mp, &tmp->tm_export, 1701 (const struct export_args *) ap->a_ctl); 1702 break; 1703 default: 1704 rc = vop_stdmountctl(ap); 1705 break; 1706 } 1707 1708 lwkt_reltoken(&mp->mnt_token); 1709 return (rc); 1710 } 1711 1712 /* --------------------------------------------------------------------- */ 1713 1714 static int 1715 tmpfs_print(struct vop_print_args *v) 1716 { 1717 struct vnode *vp = v->a_vp; 1718 1719 struct tmpfs_node *node; 1720 1721 node = VP_TO_TMPFS_NODE(vp); 1722 1723 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n", 1724 node, node->tn_flags, node->tn_links); 1725 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n", 1726 node->tn_mode, node->tn_uid, node->tn_gid, 1727 (uintmax_t)node->tn_size, node->tn_status); 1728 1729 if (vp->v_type == VFIFO) 1730 fifo_printinfo(vp); 1731 1732 kprintf("\n"); 1733 1734 return 0; 1735 } 1736 1737 /* --------------------------------------------------------------------- */ 1738 1739 static int 1740 tmpfs_pathconf(struct vop_pathconf_args *v) 1741 { 1742 int name = v->a_name; 1743 register_t *retval = v->a_retval; 1744 1745 int error; 1746 1747 error = 0; 1748 1749 switch (name) { 1750 case _PC_LINK_MAX: 1751 *retval = LINK_MAX; 1752 break; 1753 1754 case _PC_NAME_MAX: 1755 *retval = NAME_MAX; 1756 break; 1757 1758 case _PC_PATH_MAX: 1759 *retval = PATH_MAX; 1760 break; 1761 1762 case _PC_PIPE_BUF: 1763 *retval = PIPE_BUF; 1764 break; 1765 1766 case _PC_CHOWN_RESTRICTED: 1767 *retval = 1; 1768 break; 1769 1770 case _PC_NO_TRUNC: 1771 *retval = 1; 1772 break; 1773 1774 case _PC_SYNC_IO: 1775 *retval = 1; 1776 break; 1777 1778 case _PC_FILESIZEBITS: 1779 *retval = 0; /* XXX Don't know which value should I return. */ 1780 break; 1781 1782 default: 1783 error = EINVAL; 1784 } 1785 1786 return error; 1787 } 1788 1789 /************************************************************************ 1790 * KQFILTER OPS * 1791 ************************************************************************/ 1792 1793 static void filt_tmpfsdetach(struct knote *kn); 1794 static int filt_tmpfsread(struct knote *kn, long hint); 1795 static int filt_tmpfswrite(struct knote *kn, long hint); 1796 static int filt_tmpfsvnode(struct knote *kn, long hint); 1797 1798 static struct filterops tmpfsread_filtops = 1799 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread }; 1800 static struct filterops tmpfswrite_filtops = 1801 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite }; 1802 static struct filterops tmpfsvnode_filtops = 1803 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode }; 1804 1805 static int 1806 tmpfs_kqfilter (struct vop_kqfilter_args *ap) 1807 { 1808 struct vnode *vp = ap->a_vp; 1809 struct knote *kn = ap->a_kn; 1810 1811 switch (kn->kn_filter) { 1812 case EVFILT_READ: 1813 kn->kn_fop = &tmpfsread_filtops; 1814 break; 1815 case EVFILT_WRITE: 1816 kn->kn_fop = &tmpfswrite_filtops; 1817 break; 1818 case EVFILT_VNODE: 1819 kn->kn_fop = &tmpfsvnode_filtops; 1820 break; 1821 default: 1822 return (EOPNOTSUPP); 1823 } 1824 1825 kn->kn_hook = (caddr_t)vp; 1826 1827 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 1828 1829 return(0); 1830 } 1831 1832 static void 1833 filt_tmpfsdetach(struct knote *kn) 1834 { 1835 struct vnode *vp = (void *)kn->kn_hook; 1836 1837 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 1838 } 1839 1840 static int 1841 filt_tmpfsread(struct knote *kn, long hint) 1842 { 1843 struct vnode *vp = (void *)kn->kn_hook; 1844 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp); 1845 off_t off; 1846 1847 if (hint == NOTE_REVOKE) { 1848 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 1849 return(1); 1850 } 1851 1852 /* 1853 * Interlock against MP races when performing this function. 1854 */ 1855 lwkt_gettoken(&vp->v_mount->mnt_token); 1856 off = node->tn_size - kn->kn_fp->f_offset; 1857 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 1858 if (kn->kn_sfflags & NOTE_OLDAPI) { 1859 lwkt_reltoken(&vp->v_mount->mnt_token); 1860 return(1); 1861 } 1862 1863 if (kn->kn_data == 0) { 1864 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 1865 } 1866 lwkt_reltoken(&vp->v_mount->mnt_token); 1867 return (kn->kn_data != 0); 1868 } 1869 1870 static int 1871 filt_tmpfswrite(struct knote *kn, long hint) 1872 { 1873 if (hint == NOTE_REVOKE) 1874 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 1875 kn->kn_data = 0; 1876 return (1); 1877 } 1878 1879 static int 1880 filt_tmpfsvnode(struct knote *kn, long hint) 1881 { 1882 if (kn->kn_sfflags & hint) 1883 kn->kn_fflags |= hint; 1884 if (hint == NOTE_REVOKE) { 1885 kn->kn_flags |= (EV_EOF | EV_NODATA); 1886 return (1); 1887 } 1888 return (kn->kn_fflags != 0); 1889 } 1890 1891 1892 /* --------------------------------------------------------------------- */ 1893 1894 /* 1895 * vnode operations vector used for files stored in a tmpfs file system. 1896 */ 1897 struct vop_ops tmpfs_vnode_vops = { 1898 .vop_default = vop_defaultop, 1899 .vop_getpages = vop_stdgetpages, 1900 .vop_putpages = vop_stdputpages, 1901 .vop_ncreate = tmpfs_ncreate, 1902 .vop_nresolve = tmpfs_nresolve, 1903 .vop_nlookupdotdot = tmpfs_nlookupdotdot, 1904 .vop_nmknod = tmpfs_nmknod, 1905 .vop_open = tmpfs_open, 1906 .vop_close = tmpfs_close, 1907 .vop_access = tmpfs_access, 1908 .vop_getattr = tmpfs_getattr, 1909 .vop_setattr = tmpfs_setattr, 1910 .vop_read = tmpfs_read, 1911 .vop_write = tmpfs_write, 1912 .vop_fsync = tmpfs_fsync, 1913 .vop_mountctl = tmpfs_mountctl, 1914 .vop_nremove = tmpfs_nremove, 1915 .vop_nlink = tmpfs_nlink, 1916 .vop_nrename = tmpfs_nrename, 1917 .vop_nmkdir = tmpfs_nmkdir, 1918 .vop_nrmdir = tmpfs_nrmdir, 1919 .vop_nsymlink = tmpfs_nsymlink, 1920 .vop_readdir = tmpfs_readdir, 1921 .vop_readlink = tmpfs_readlink, 1922 .vop_inactive = tmpfs_inactive, 1923 .vop_reclaim = tmpfs_reclaim, 1924 .vop_print = tmpfs_print, 1925 .vop_pathconf = tmpfs_pathconf, 1926 .vop_bmap = tmpfs_bmap, 1927 .vop_strategy = tmpfs_strategy, 1928 .vop_advlock = tmpfs_advlock, 1929 .vop_kqfilter = tmpfs_kqfilter 1930 }; 1931