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