1 /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */ 2 3 /*- 4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD 5 * 6 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc. 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to The NetBSD Foundation 10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 11 * 2005 program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 32 * POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 /* 36 * tmpfs vnode interface. 37 */ 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/dirent.h> 44 #include <sys/fcntl.h> 45 #include <sys/file.h> 46 #include <sys/filio.h> 47 #include <sys/limits.h> 48 #include <sys/lockf.h> 49 #include <sys/lock.h> 50 #include <sys/mount.h> 51 #include <sys/namei.h> 52 #include <sys/priv.h> 53 #include <sys/proc.h> 54 #include <sys/rwlock.h> 55 #include <sys/sched.h> 56 #include <sys/smr.h> 57 #include <sys/stat.h> 58 #include <sys/sysctl.h> 59 #include <sys/unistd.h> 60 #include <sys/vnode.h> 61 #include <security/audit/audit.h> 62 #include <security/mac/mac_framework.h> 63 64 #include <vm/vm.h> 65 #include <vm/vm_param.h> 66 #include <vm/vm_object.h> 67 #include <vm/vm_page.h> 68 #include <vm/vm_pager.h> 69 #include <vm/swap_pager.h> 70 71 #include <fs/tmpfs/tmpfs_vnops.h> 72 #include <fs/tmpfs/tmpfs.h> 73 74 SYSCTL_DECL(_vfs_tmpfs); 75 VFS_SMR_DECLARE; 76 77 static volatile int tmpfs_rename_restarts; 78 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD, 79 __DEVOLATILE(int *, &tmpfs_rename_restarts), 0, 80 "Times rename had to restart due to lock contention"); 81 82 static int 83 tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags, 84 struct vnode **rvp) 85 { 86 87 return (tmpfs_alloc_vp(mp, arg, lkflags, rvp)); 88 } 89 90 static int 91 tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 92 { 93 struct tmpfs_dirent *de; 94 struct tmpfs_node *dnode, *pnode; 95 struct tmpfs_mount *tm; 96 int error; 97 98 /* Caller assumes responsibility for ensuring access (VEXEC). */ 99 dnode = VP_TO_TMPFS_DIR(dvp); 100 *vpp = NULLVP; 101 102 /* We cannot be requesting the parent directory of the root node. */ 103 MPASS(IMPLIES(dnode->tn_type == VDIR && 104 dnode->tn_dir.tn_parent == dnode, 105 !(cnp->cn_flags & ISDOTDOT))); 106 107 TMPFS_ASSERT_LOCKED(dnode); 108 if (dnode->tn_dir.tn_parent == NULL) { 109 error = ENOENT; 110 goto out; 111 } 112 if (cnp->cn_flags & ISDOTDOT) { 113 tm = VFS_TO_TMPFS(dvp->v_mount); 114 pnode = dnode->tn_dir.tn_parent; 115 tmpfs_ref_node(pnode); 116 error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc, 117 pnode, cnp->cn_lkflags, vpp); 118 tmpfs_free_node(tm, pnode); 119 if (error != 0) 120 goto out; 121 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 122 VREF(dvp); 123 *vpp = dvp; 124 error = 0; 125 } else { 126 de = tmpfs_dir_lookup(dnode, NULL, cnp); 127 if (de != NULL && de->td_node == NULL) 128 cnp->cn_flags |= ISWHITEOUT; 129 if (de == NULL || de->td_node == NULL) { 130 /* 131 * The entry was not found in the directory. 132 * This is OK if we are creating or renaming an 133 * entry and are working on the last component of 134 * the path name. 135 */ 136 if ((cnp->cn_flags & ISLASTCN) && 137 (cnp->cn_nameiop == CREATE || \ 138 cnp->cn_nameiop == RENAME || 139 (cnp->cn_nameiop == DELETE && 140 cnp->cn_flags & DOWHITEOUT && 141 cnp->cn_flags & ISWHITEOUT))) { 142 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 143 curthread); 144 if (error != 0) 145 goto out; 146 147 error = EJUSTRETURN; 148 } else 149 error = ENOENT; 150 } else { 151 struct tmpfs_node *tnode; 152 153 /* 154 * The entry was found, so get its associated 155 * tmpfs_node. 156 */ 157 tnode = de->td_node; 158 159 /* 160 * If we are not at the last path component and 161 * found a non-directory or non-link entry (which 162 * may itself be pointing to a directory), raise 163 * an error. 164 */ 165 if ((tnode->tn_type != VDIR && 166 tnode->tn_type != VLNK) && 167 !(cnp->cn_flags & ISLASTCN)) { 168 error = ENOTDIR; 169 goto out; 170 } 171 172 /* 173 * If we are deleting or renaming the entry, keep 174 * track of its tmpfs_dirent so that it can be 175 * easily deleted later. 176 */ 177 if ((cnp->cn_flags & ISLASTCN) && 178 (cnp->cn_nameiop == DELETE || 179 cnp->cn_nameiop == RENAME)) { 180 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 181 curthread); 182 if (error != 0) 183 goto out; 184 185 /* Allocate a new vnode on the matching entry. */ 186 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 187 cnp->cn_lkflags, vpp); 188 if (error != 0) 189 goto out; 190 191 if ((dnode->tn_mode & S_ISTXT) && 192 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, 193 curthread) && VOP_ACCESS(*vpp, VADMIN, 194 cnp->cn_cred, curthread)) { 195 error = EPERM; 196 vput(*vpp); 197 *vpp = NULL; 198 goto out; 199 } 200 } else { 201 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 202 cnp->cn_lkflags, vpp); 203 if (error != 0) 204 goto out; 205 } 206 } 207 } 208 209 /* 210 * Store the result of this lookup in the cache. Avoid this if the 211 * request was for creation, as it does not improve timings on 212 * emprical tests. 213 */ 214 if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) 215 cache_enter(dvp, *vpp, cnp); 216 217 out: 218 /* 219 * If there were no errors, *vpp cannot be null and it must be 220 * locked. 221 */ 222 MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp))); 223 224 return (error); 225 } 226 227 static int 228 tmpfs_cached_lookup(struct vop_cachedlookup_args *v) 229 { 230 231 return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp)); 232 } 233 234 static int 235 tmpfs_lookup(struct vop_lookup_args *v) 236 { 237 struct vnode *dvp = v->a_dvp; 238 struct vnode **vpp = v->a_vpp; 239 struct componentname *cnp = v->a_cnp; 240 int error; 241 242 /* Check accessibility of requested node as a first step. */ 243 error = vn_dir_check_exec(dvp, cnp); 244 if (error != 0) 245 return (error); 246 247 return (tmpfs_lookup1(dvp, vpp, cnp)); 248 } 249 250 static int 251 tmpfs_create(struct vop_create_args *v) 252 { 253 struct vnode *dvp = v->a_dvp; 254 struct vnode **vpp = v->a_vpp; 255 struct componentname *cnp = v->a_cnp; 256 struct vattr *vap = v->a_vap; 257 int error; 258 259 MPASS(vap->va_type == VREG || vap->va_type == VSOCK); 260 261 error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 262 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) 263 cache_enter(dvp, *vpp, cnp); 264 return (error); 265 } 266 267 static int 268 tmpfs_mknod(struct vop_mknod_args *v) 269 { 270 struct vnode *dvp = v->a_dvp; 271 struct vnode **vpp = v->a_vpp; 272 struct componentname *cnp = v->a_cnp; 273 struct vattr *vap = v->a_vap; 274 275 if (vap->va_type != VBLK && vap->va_type != VCHR && 276 vap->va_type != VFIFO) 277 return (EINVAL); 278 279 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL)); 280 } 281 282 struct fileops tmpfs_fnops; 283 284 static int 285 tmpfs_open(struct vop_open_args *v) 286 { 287 struct vnode *vp; 288 struct tmpfs_node *node; 289 struct file *fp; 290 int error, mode; 291 292 vp = v->a_vp; 293 mode = v->a_mode; 294 node = VP_TO_TMPFS_NODE(vp); 295 296 /* 297 * The file is still active but all its names have been removed 298 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 299 * it is about to die. 300 */ 301 if (node->tn_links < 1) 302 return (ENOENT); 303 304 /* If the file is marked append-only, deny write requests. */ 305 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE) 306 error = EPERM; 307 else { 308 error = 0; 309 /* For regular files, the call below is nop. */ 310 KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags & 311 OBJ_DEAD) == 0, ("dead object")); 312 vnode_create_vobject(vp, node->tn_size, v->a_td); 313 } 314 315 fp = v->a_fp; 316 MPASS(fp == NULL || fp->f_data == NULL); 317 if (error == 0 && fp != NULL && vp->v_type == VREG) { 318 tmpfs_ref_node(node); 319 finit_vnode(fp, mode, node, &tmpfs_fnops); 320 } 321 322 return (error); 323 } 324 325 static int 326 tmpfs_close(struct vop_close_args *v) 327 { 328 struct vnode *vp = v->a_vp; 329 330 /* Update node times. */ 331 tmpfs_update(vp); 332 333 return (0); 334 } 335 336 int 337 tmpfs_fo_close(struct file *fp, struct thread *td) 338 { 339 struct tmpfs_node *node; 340 341 node = fp->f_data; 342 if (node != NULL) { 343 MPASS(node->tn_type == VREG); 344 tmpfs_free_node(node->tn_reg.tn_tmp, node); 345 } 346 return (vnops.fo_close(fp, td)); 347 } 348 349 /* 350 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see 351 * the comment above cache_fplookup for details. 352 */ 353 int 354 tmpfs_fplookup_vexec(struct vop_fplookup_vexec_args *v) 355 { 356 struct vnode *vp; 357 struct tmpfs_node *node; 358 struct ucred *cred; 359 mode_t all_x, mode; 360 361 vp = v->a_vp; 362 node = VP_TO_TMPFS_NODE_SMR(vp); 363 if (__predict_false(node == NULL)) 364 return (EAGAIN); 365 366 all_x = S_IXUSR | S_IXGRP | S_IXOTH; 367 mode = atomic_load_short(&node->tn_mode); 368 if (__predict_true((mode & all_x) == all_x)) 369 return (0); 370 371 cred = v->a_cred; 372 return (vaccess_vexec_smr(mode, node->tn_uid, node->tn_gid, cred)); 373 } 374 375 int 376 tmpfs_access(struct vop_access_args *v) 377 { 378 struct vnode *vp = v->a_vp; 379 accmode_t accmode = v->a_accmode; 380 struct ucred *cred = v->a_cred; 381 mode_t all_x = S_IXUSR | S_IXGRP | S_IXOTH; 382 int error; 383 struct tmpfs_node *node; 384 385 MPASS(VOP_ISLOCKED(vp)); 386 387 node = VP_TO_TMPFS_NODE(vp); 388 389 /* 390 * Common case path lookup. 391 */ 392 if (__predict_true(accmode == VEXEC && (node->tn_mode & all_x) == all_x)) 393 return (0); 394 395 switch (vp->v_type) { 396 case VDIR: 397 /* FALLTHROUGH */ 398 case VLNK: 399 /* FALLTHROUGH */ 400 case VREG: 401 if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) { 402 error = EROFS; 403 goto out; 404 } 405 break; 406 407 case VBLK: 408 /* FALLTHROUGH */ 409 case VCHR: 410 /* FALLTHROUGH */ 411 case VSOCK: 412 /* FALLTHROUGH */ 413 case VFIFO: 414 break; 415 416 default: 417 error = EINVAL; 418 goto out; 419 } 420 421 if (accmode & VWRITE && node->tn_flags & IMMUTABLE) { 422 error = EPERM; 423 goto out; 424 } 425 426 error = vaccess(vp->v_type, node->tn_mode, node->tn_uid, node->tn_gid, 427 accmode, cred); 428 429 out: 430 MPASS(VOP_ISLOCKED(vp)); 431 432 return (error); 433 } 434 435 int 436 tmpfs_stat(struct vop_stat_args *v) 437 { 438 struct vnode *vp = v->a_vp; 439 struct stat *sb = v->a_sb; 440 vm_object_t obj; 441 struct tmpfs_node *node; 442 int error; 443 444 node = VP_TO_TMPFS_NODE(vp); 445 446 tmpfs_update_getattr(vp); 447 448 error = vop_stat_helper_pre(v); 449 if (__predict_false(error)) 450 return (error); 451 452 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 453 sb->st_ino = node->tn_id; 454 sb->st_mode = node->tn_mode | VTTOIF(vp->v_type); 455 sb->st_nlink = node->tn_links; 456 sb->st_uid = node->tn_uid; 457 sb->st_gid = node->tn_gid; 458 sb->st_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? 459 node->tn_rdev : NODEV; 460 sb->st_size = node->tn_size; 461 sb->st_atim.tv_sec = node->tn_atime.tv_sec; 462 sb->st_atim.tv_nsec = node->tn_atime.tv_nsec; 463 sb->st_mtim.tv_sec = node->tn_mtime.tv_sec; 464 sb->st_mtim.tv_nsec = node->tn_mtime.tv_nsec; 465 sb->st_ctim.tv_sec = node->tn_ctime.tv_sec; 466 sb->st_ctim.tv_nsec = node->tn_ctime.tv_nsec; 467 sb->st_birthtim.tv_sec = node->tn_birthtime.tv_sec; 468 sb->st_birthtim.tv_nsec = node->tn_birthtime.tv_nsec; 469 sb->st_blksize = PAGE_SIZE; 470 sb->st_flags = node->tn_flags; 471 sb->st_gen = node->tn_gen; 472 if (vp->v_type == VREG) { 473 obj = node->tn_reg.tn_aobj; 474 sb->st_blocks = (u_quad_t)obj->resident_page_count * PAGE_SIZE; 475 } else 476 sb->st_blocks = node->tn_size; 477 sb->st_blocks /= S_BLKSIZE; 478 return (vop_stat_helper_post(v, error)); 479 } 480 481 int 482 tmpfs_getattr(struct vop_getattr_args *v) 483 { 484 struct vnode *vp = v->a_vp; 485 struct vattr *vap = v->a_vap; 486 vm_object_t obj; 487 struct tmpfs_node *node; 488 489 node = VP_TO_TMPFS_NODE(vp); 490 491 tmpfs_update_getattr(vp); 492 493 vap->va_type = vp->v_type; 494 vap->va_mode = node->tn_mode; 495 vap->va_nlink = node->tn_links; 496 vap->va_uid = node->tn_uid; 497 vap->va_gid = node->tn_gid; 498 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 499 vap->va_fileid = node->tn_id; 500 vap->va_size = node->tn_size; 501 vap->va_blocksize = PAGE_SIZE; 502 vap->va_atime = node->tn_atime; 503 vap->va_mtime = node->tn_mtime; 504 vap->va_ctime = node->tn_ctime; 505 vap->va_birthtime = node->tn_birthtime; 506 vap->va_gen = node->tn_gen; 507 vap->va_flags = node->tn_flags; 508 vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? 509 node->tn_rdev : NODEV; 510 if (vp->v_type == VREG) { 511 obj = node->tn_reg.tn_aobj; 512 vap->va_bytes = (u_quad_t)obj->resident_page_count * PAGE_SIZE; 513 } else 514 vap->va_bytes = node->tn_size; 515 vap->va_filerev = 0; 516 517 return (0); 518 } 519 520 int 521 tmpfs_setattr(struct vop_setattr_args *v) 522 { 523 struct vnode *vp = v->a_vp; 524 struct vattr *vap = v->a_vap; 525 struct ucred *cred = v->a_cred; 526 struct thread *td = curthread; 527 528 int error; 529 530 MPASS(VOP_ISLOCKED(vp)); 531 ASSERT_VOP_IN_SEQC(vp); 532 533 error = 0; 534 535 /* Abort if any unsettable attribute is given. */ 536 if (vap->va_type != VNON || 537 vap->va_nlink != VNOVAL || 538 vap->va_fsid != VNOVAL || 539 vap->va_fileid != VNOVAL || 540 vap->va_blocksize != VNOVAL || 541 vap->va_gen != VNOVAL || 542 vap->va_rdev != VNOVAL || 543 vap->va_bytes != VNOVAL) 544 error = EINVAL; 545 546 if (error == 0 && (vap->va_flags != VNOVAL)) 547 error = tmpfs_chflags(vp, vap->va_flags, cred, td); 548 549 if (error == 0 && (vap->va_size != VNOVAL)) 550 error = tmpfs_chsize(vp, vap->va_size, cred, td); 551 552 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL)) 553 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td); 554 555 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) 556 error = tmpfs_chmod(vp, vap->va_mode, cred, td); 557 558 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && 559 vap->va_atime.tv_nsec != VNOVAL) || 560 (vap->va_mtime.tv_sec != VNOVAL && 561 vap->va_mtime.tv_nsec != VNOVAL) || 562 (vap->va_birthtime.tv_sec != VNOVAL && 563 vap->va_birthtime.tv_nsec != VNOVAL))) 564 error = tmpfs_chtimes(vp, vap, cred, td); 565 566 /* 567 * Update the node times. We give preference to the error codes 568 * generated by this function rather than the ones that may arise 569 * from tmpfs_update. 570 */ 571 tmpfs_update(vp); 572 573 MPASS(VOP_ISLOCKED(vp)); 574 575 return (error); 576 } 577 578 static int 579 tmpfs_read(struct vop_read_args *v) 580 { 581 struct vnode *vp; 582 struct uio *uio; 583 struct tmpfs_node *node; 584 585 vp = v->a_vp; 586 if (vp->v_type != VREG) 587 return (EISDIR); 588 uio = v->a_uio; 589 if (uio->uio_offset < 0) 590 return (EINVAL); 591 node = VP_TO_TMPFS_NODE(vp); 592 tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node); 593 return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio)); 594 } 595 596 static int 597 tmpfs_read_pgcache(struct vop_read_pgcache_args *v) 598 { 599 struct vnode *vp; 600 struct tmpfs_node *node; 601 vm_object_t object; 602 off_t size; 603 int error; 604 605 vp = v->a_vp; 606 VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp); 607 608 if (v->a_uio->uio_offset < 0) 609 return (EINVAL); 610 611 error = EJUSTRETURN; 612 vfs_smr_enter(); 613 614 node = VP_TO_TMPFS_NODE_SMR(vp); 615 if (node == NULL) 616 goto out_smr; 617 MPASS(node->tn_type == VREG); 618 MPASS(node->tn_refcount >= 1); 619 object = node->tn_reg.tn_aobj; 620 if (object == NULL) 621 goto out_smr; 622 623 MPASS(object->type == tmpfs_pager_type); 624 MPASS((object->flags & (OBJ_ANON | OBJ_DEAD | OBJ_SWAP)) == 625 OBJ_SWAP); 626 if (!VN_IS_DOOMED(vp)) { 627 /* size cannot become shorter due to rangelock. */ 628 size = node->tn_size; 629 tmpfs_set_accessed(node->tn_reg.tn_tmp, node); 630 vfs_smr_exit(); 631 error = uiomove_object(object, size, v->a_uio); 632 return (error); 633 } 634 out_smr: 635 vfs_smr_exit(); 636 return (error); 637 } 638 639 static int 640 tmpfs_write(struct vop_write_args *v) 641 { 642 struct vnode *vp; 643 struct uio *uio; 644 struct tmpfs_node *node; 645 off_t oldsize; 646 ssize_t r; 647 int error, ioflag; 648 mode_t newmode; 649 650 vp = v->a_vp; 651 uio = v->a_uio; 652 ioflag = v->a_ioflag; 653 error = 0; 654 node = VP_TO_TMPFS_NODE(vp); 655 oldsize = node->tn_size; 656 657 if (uio->uio_offset < 0 || vp->v_type != VREG) 658 return (EINVAL); 659 if (uio->uio_resid == 0) 660 return (0); 661 if (ioflag & IO_APPEND) 662 uio->uio_offset = node->tn_size; 663 error = vn_rlimit_fsizex(vp, uio, VFS_TO_TMPFS(vp->v_mount)-> 664 tm_maxfilesize, &r, uio->uio_td); 665 if (error != 0) { 666 vn_rlimit_fsizex_res(uio, r); 667 return (error); 668 } 669 670 if (uio->uio_offset + uio->uio_resid > node->tn_size) { 671 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid, 672 FALSE); 673 if (error != 0) 674 goto out; 675 } 676 677 error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio); 678 node->tn_status |= TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED; 679 node->tn_accessed = true; 680 if (node->tn_mode & (S_ISUID | S_ISGID)) { 681 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID)) { 682 newmode = node->tn_mode & ~(S_ISUID | S_ISGID); 683 vn_seqc_write_begin(vp); 684 atomic_store_short(&node->tn_mode, newmode); 685 vn_seqc_write_end(vp); 686 } 687 } 688 if (error != 0) 689 (void)tmpfs_reg_resize(vp, oldsize, TRUE); 690 691 out: 692 MPASS(IMPLIES(error == 0, uio->uio_resid == 0)); 693 MPASS(IMPLIES(error != 0, oldsize == node->tn_size)); 694 695 vn_rlimit_fsizex_res(uio, r); 696 return (error); 697 } 698 699 static int 700 tmpfs_deallocate(struct vop_deallocate_args *v) 701 { 702 return (tmpfs_reg_punch_hole(v->a_vp, v->a_offset, v->a_len)); 703 } 704 705 static int 706 tmpfs_fsync(struct vop_fsync_args *v) 707 { 708 struct vnode *vp = v->a_vp; 709 710 MPASS(VOP_ISLOCKED(vp)); 711 712 tmpfs_check_mtime(vp); 713 tmpfs_update(vp); 714 715 return (0); 716 } 717 718 static int 719 tmpfs_remove(struct vop_remove_args *v) 720 { 721 struct vnode *dvp = v->a_dvp; 722 struct vnode *vp = v->a_vp; 723 724 int error; 725 struct tmpfs_dirent *de; 726 struct tmpfs_mount *tmp; 727 struct tmpfs_node *dnode; 728 struct tmpfs_node *node; 729 730 MPASS(VOP_ISLOCKED(dvp)); 731 MPASS(VOP_ISLOCKED(vp)); 732 733 if (vp->v_type == VDIR) { 734 error = EISDIR; 735 goto out; 736 } 737 738 dnode = VP_TO_TMPFS_DIR(dvp); 739 node = VP_TO_TMPFS_NODE(vp); 740 tmp = VFS_TO_TMPFS(vp->v_mount); 741 de = tmpfs_dir_lookup(dnode, node, v->a_cnp); 742 MPASS(de != NULL); 743 744 /* Files marked as immutable or append-only cannot be deleted. */ 745 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 746 (dnode->tn_flags & APPEND)) { 747 error = EPERM; 748 goto out; 749 } 750 751 /* Remove the entry from the directory; as it is a file, we do not 752 * have to change the number of hard links of the directory. */ 753 tmpfs_dir_detach(dvp, de); 754 if (v->a_cnp->cn_flags & DOWHITEOUT) 755 tmpfs_dir_whiteout_add(dvp, v->a_cnp); 756 757 /* Free the directory entry we just deleted. Note that the node 758 * referred by it will not be removed until the vnode is really 759 * reclaimed. */ 760 tmpfs_free_dirent(tmp, de); 761 762 node->tn_status |= TMPFS_NODE_CHANGED; 763 node->tn_accessed = true; 764 error = 0; 765 766 out: 767 return (error); 768 } 769 770 static int 771 tmpfs_link(struct vop_link_args *v) 772 { 773 struct vnode *dvp = v->a_tdvp; 774 struct vnode *vp = v->a_vp; 775 struct componentname *cnp = v->a_cnp; 776 777 int error; 778 struct tmpfs_dirent *de; 779 struct tmpfs_node *node; 780 781 MPASS(VOP_ISLOCKED(dvp)); 782 MPASS(dvp != vp); /* XXX When can this be false? */ 783 node = VP_TO_TMPFS_NODE(vp); 784 785 /* Ensure that we do not overflow the maximum number of links imposed 786 * by the system. */ 787 MPASS(node->tn_links <= TMPFS_LINK_MAX); 788 if (node->tn_links == TMPFS_LINK_MAX) { 789 error = EMLINK; 790 goto out; 791 } 792 793 /* We cannot create links of files marked immutable or append-only. */ 794 if (node->tn_flags & (IMMUTABLE | APPEND)) { 795 error = EPERM; 796 goto out; 797 } 798 799 /* Allocate a new directory entry to represent the node. */ 800 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 801 cnp->cn_nameptr, cnp->cn_namelen, &de); 802 if (error != 0) 803 goto out; 804 805 /* Insert the new directory entry into the appropriate directory. */ 806 if (cnp->cn_flags & ISWHITEOUT) 807 tmpfs_dir_whiteout_remove(dvp, cnp); 808 tmpfs_dir_attach(dvp, de); 809 810 /* vp link count has changed, so update node times. */ 811 node->tn_status |= TMPFS_NODE_CHANGED; 812 tmpfs_update(vp); 813 814 error = 0; 815 816 out: 817 return (error); 818 } 819 820 /* 821 * We acquire all but fdvp locks using non-blocking acquisitions. If we 822 * fail to acquire any lock in the path we will drop all held locks, 823 * acquire the new lock in a blocking fashion, and then release it and 824 * restart the rename. This acquire/release step ensures that we do not 825 * spin on a lock waiting for release. On error release all vnode locks 826 * and decrement references the way tmpfs_rename() would do. 827 */ 828 static int 829 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp, 830 struct vnode *tdvp, struct vnode **tvpp, 831 struct componentname *fcnp, struct componentname *tcnp) 832 { 833 struct vnode *nvp; 834 struct mount *mp; 835 struct tmpfs_dirent *de; 836 int error, restarts = 0; 837 838 VOP_UNLOCK(tdvp); 839 if (*tvpp != NULL && *tvpp != tdvp) 840 VOP_UNLOCK(*tvpp); 841 mp = fdvp->v_mount; 842 843 relock: 844 restarts += 1; 845 error = vn_lock(fdvp, LK_EXCLUSIVE); 846 if (error) 847 goto releout; 848 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 849 VOP_UNLOCK(fdvp); 850 error = vn_lock(tdvp, LK_EXCLUSIVE); 851 if (error) 852 goto releout; 853 VOP_UNLOCK(tdvp); 854 goto relock; 855 } 856 /* 857 * Re-resolve fvp to be certain it still exists and fetch the 858 * correct vnode. 859 */ 860 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp); 861 if (de == NULL) { 862 VOP_UNLOCK(fdvp); 863 VOP_UNLOCK(tdvp); 864 if ((fcnp->cn_flags & ISDOTDOT) != 0 || 865 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) 866 error = EINVAL; 867 else 868 error = ENOENT; 869 goto releout; 870 } 871 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp); 872 if (error != 0) { 873 VOP_UNLOCK(fdvp); 874 VOP_UNLOCK(tdvp); 875 if (error != EBUSY) 876 goto releout; 877 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp); 878 if (error != 0) 879 goto releout; 880 VOP_UNLOCK(nvp); 881 /* 882 * Concurrent rename race. 883 */ 884 if (nvp == tdvp) { 885 vrele(nvp); 886 error = EINVAL; 887 goto releout; 888 } 889 vrele(*fvpp); 890 *fvpp = nvp; 891 goto relock; 892 } 893 vrele(*fvpp); 894 *fvpp = nvp; 895 VOP_UNLOCK(*fvpp); 896 /* 897 * Re-resolve tvp and acquire the vnode lock if present. 898 */ 899 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp); 900 /* 901 * If tvp disappeared we just carry on. 902 */ 903 if (de == NULL && *tvpp != NULL) { 904 vrele(*tvpp); 905 *tvpp = NULL; 906 } 907 /* 908 * Get the tvp ino if the lookup succeeded. We may have to restart 909 * if the non-blocking acquire fails. 910 */ 911 if (de != NULL) { 912 nvp = NULL; 913 error = tmpfs_alloc_vp(mp, de->td_node, 914 LK_EXCLUSIVE | LK_NOWAIT, &nvp); 915 if (*tvpp != NULL) 916 vrele(*tvpp); 917 *tvpp = nvp; 918 if (error != 0) { 919 VOP_UNLOCK(fdvp); 920 VOP_UNLOCK(tdvp); 921 if (error != EBUSY) 922 goto releout; 923 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, 924 &nvp); 925 if (error != 0) 926 goto releout; 927 VOP_UNLOCK(nvp); 928 /* 929 * fdvp contains fvp, thus tvp (=fdvp) is not empty. 930 */ 931 if (nvp == fdvp) { 932 error = ENOTEMPTY; 933 goto releout; 934 } 935 goto relock; 936 } 937 } 938 tmpfs_rename_restarts += restarts; 939 940 return (0); 941 942 releout: 943 vrele(fdvp); 944 vrele(*fvpp); 945 vrele(tdvp); 946 if (*tvpp != NULL) 947 vrele(*tvpp); 948 tmpfs_rename_restarts += restarts; 949 950 return (error); 951 } 952 953 static int 954 tmpfs_rename(struct vop_rename_args *v) 955 { 956 struct vnode *fdvp = v->a_fdvp; 957 struct vnode *fvp = v->a_fvp; 958 struct componentname *fcnp = v->a_fcnp; 959 struct vnode *tdvp = v->a_tdvp; 960 struct vnode *tvp = v->a_tvp; 961 struct componentname *tcnp = v->a_tcnp; 962 char *newname; 963 struct tmpfs_dirent *de; 964 struct tmpfs_mount *tmp; 965 struct tmpfs_node *fdnode; 966 struct tmpfs_node *fnode; 967 struct tmpfs_node *tnode; 968 struct tmpfs_node *tdnode; 969 int error; 970 bool want_seqc_end; 971 972 MPASS(VOP_ISLOCKED(tdvp)); 973 MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp))); 974 975 want_seqc_end = false; 976 977 /* 978 * Disallow cross-device renames. 979 * XXX Why isn't this done by the caller? 980 */ 981 if (fvp->v_mount != tdvp->v_mount || 982 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 983 error = EXDEV; 984 goto out; 985 } 986 987 /* If source and target are the same file, there is nothing to do. */ 988 if (fvp == tvp) { 989 error = 0; 990 goto out; 991 } 992 993 /* 994 * If we need to move the directory between entries, lock the 995 * source so that we can safely operate on it. 996 */ 997 if (fdvp != tdvp && fdvp != tvp) { 998 if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 999 error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp, 1000 fcnp, tcnp); 1001 if (error != 0) 1002 return (error); 1003 ASSERT_VOP_ELOCKED(fdvp, 1004 "tmpfs_rename: fdvp not locked"); 1005 ASSERT_VOP_ELOCKED(tdvp, 1006 "tmpfs_rename: tdvp not locked"); 1007 if (tvp != NULL) 1008 ASSERT_VOP_ELOCKED(tvp, 1009 "tmpfs_rename: tvp not locked"); 1010 if (fvp == tvp) { 1011 error = 0; 1012 goto out_locked; 1013 } 1014 } 1015 } 1016 1017 if (tvp != NULL) 1018 vn_seqc_write_begin(tvp); 1019 vn_seqc_write_begin(tdvp); 1020 vn_seqc_write_begin(fvp); 1021 vn_seqc_write_begin(fdvp); 1022 want_seqc_end = true; 1023 1024 tmp = VFS_TO_TMPFS(tdvp->v_mount); 1025 tdnode = VP_TO_TMPFS_DIR(tdvp); 1026 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp); 1027 fdnode = VP_TO_TMPFS_DIR(fdvp); 1028 fnode = VP_TO_TMPFS_NODE(fvp); 1029 de = tmpfs_dir_lookup(fdnode, fnode, fcnp); 1030 1031 /* 1032 * Entry can disappear before we lock fdvp, 1033 * also avoid manipulating '.' and '..' entries. 1034 */ 1035 if (de == NULL) { 1036 if ((fcnp->cn_flags & ISDOTDOT) != 0 || 1037 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) 1038 error = EINVAL; 1039 else 1040 error = ENOENT; 1041 goto out_locked; 1042 } 1043 MPASS(de->td_node == fnode); 1044 1045 /* 1046 * If re-naming a directory to another preexisting directory 1047 * ensure that the target directory is empty so that its 1048 * removal causes no side effects. 1049 * Kern_rename guarantees the destination to be a directory 1050 * if the source is one. 1051 */ 1052 if (tvp != NULL) { 1053 MPASS(tnode != NULL); 1054 1055 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1056 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 1057 error = EPERM; 1058 goto out_locked; 1059 } 1060 1061 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { 1062 if (tnode->tn_size > 0) { 1063 error = ENOTEMPTY; 1064 goto out_locked; 1065 } 1066 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { 1067 error = ENOTDIR; 1068 goto out_locked; 1069 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { 1070 error = EISDIR; 1071 goto out_locked; 1072 } else { 1073 MPASS(fnode->tn_type != VDIR && 1074 tnode->tn_type != VDIR); 1075 } 1076 } 1077 1078 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) 1079 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 1080 error = EPERM; 1081 goto out_locked; 1082 } 1083 1084 /* 1085 * Ensure that we have enough memory to hold the new name, if it 1086 * has to be changed. 1087 */ 1088 if (fcnp->cn_namelen != tcnp->cn_namelen || 1089 bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) { 1090 newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK); 1091 } else 1092 newname = NULL; 1093 1094 /* 1095 * If the node is being moved to another directory, we have to do 1096 * the move. 1097 */ 1098 if (fdnode != tdnode) { 1099 /* 1100 * In case we are moving a directory, we have to adjust its 1101 * parent to point to the new parent. 1102 */ 1103 if (de->td_node->tn_type == VDIR) { 1104 struct tmpfs_node *n; 1105 1106 /* 1107 * Ensure the target directory is not a child of the 1108 * directory being moved. Otherwise, we'd end up 1109 * with stale nodes. 1110 */ 1111 n = tdnode; 1112 /* 1113 * TMPFS_LOCK guaranties that no nodes are freed while 1114 * traversing the list. Nodes can only be marked as 1115 * removed: tn_parent == NULL. 1116 */ 1117 TMPFS_LOCK(tmp); 1118 TMPFS_NODE_LOCK(n); 1119 while (n != n->tn_dir.tn_parent) { 1120 struct tmpfs_node *parent; 1121 1122 if (n == fnode) { 1123 TMPFS_NODE_UNLOCK(n); 1124 TMPFS_UNLOCK(tmp); 1125 error = EINVAL; 1126 if (newname != NULL) 1127 free(newname, M_TMPFSNAME); 1128 goto out_locked; 1129 } 1130 parent = n->tn_dir.tn_parent; 1131 TMPFS_NODE_UNLOCK(n); 1132 if (parent == NULL) { 1133 n = NULL; 1134 break; 1135 } 1136 TMPFS_NODE_LOCK(parent); 1137 if (parent->tn_dir.tn_parent == NULL) { 1138 TMPFS_NODE_UNLOCK(parent); 1139 n = NULL; 1140 break; 1141 } 1142 n = parent; 1143 } 1144 TMPFS_UNLOCK(tmp); 1145 if (n == NULL) { 1146 error = EINVAL; 1147 if (newname != NULL) 1148 free(newname, M_TMPFSNAME); 1149 goto out_locked; 1150 } 1151 TMPFS_NODE_UNLOCK(n); 1152 1153 /* Adjust the parent pointer. */ 1154 TMPFS_VALIDATE_DIR(fnode); 1155 TMPFS_NODE_LOCK(de->td_node); 1156 de->td_node->tn_dir.tn_parent = tdnode; 1157 TMPFS_NODE_UNLOCK(de->td_node); 1158 1159 /* 1160 * As a result of changing the target of the '..' 1161 * entry, the link count of the source and target 1162 * directories has to be adjusted. 1163 */ 1164 TMPFS_NODE_LOCK(tdnode); 1165 TMPFS_ASSERT_LOCKED(tdnode); 1166 tdnode->tn_links++; 1167 TMPFS_NODE_UNLOCK(tdnode); 1168 1169 TMPFS_NODE_LOCK(fdnode); 1170 TMPFS_ASSERT_LOCKED(fdnode); 1171 fdnode->tn_links--; 1172 TMPFS_NODE_UNLOCK(fdnode); 1173 } 1174 } 1175 1176 /* 1177 * Do the move: just remove the entry from the source directory 1178 * and insert it into the target one. 1179 */ 1180 tmpfs_dir_detach(fdvp, de); 1181 1182 if (fcnp->cn_flags & DOWHITEOUT) 1183 tmpfs_dir_whiteout_add(fdvp, fcnp); 1184 if (tcnp->cn_flags & ISWHITEOUT) 1185 tmpfs_dir_whiteout_remove(tdvp, tcnp); 1186 1187 /* 1188 * If the name has changed, we need to make it effective by changing 1189 * it in the directory entry. 1190 */ 1191 if (newname != NULL) { 1192 MPASS(tcnp->cn_namelen <= MAXNAMLEN); 1193 1194 free(de->ud.td_name, M_TMPFSNAME); 1195 de->ud.td_name = newname; 1196 tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen); 1197 1198 fnode->tn_status |= TMPFS_NODE_CHANGED; 1199 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 1200 } 1201 1202 /* 1203 * If we are overwriting an entry, we have to remove the old one 1204 * from the target directory. 1205 */ 1206 if (tvp != NULL) { 1207 struct tmpfs_dirent *tde; 1208 1209 /* Remove the old entry from the target directory. */ 1210 tde = tmpfs_dir_lookup(tdnode, tnode, tcnp); 1211 tmpfs_dir_detach(tdvp, tde); 1212 1213 /* 1214 * Free the directory entry we just deleted. Note that the 1215 * node referred by it will not be removed until the vnode is 1216 * really reclaimed. 1217 */ 1218 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde); 1219 } 1220 1221 tmpfs_dir_attach(tdvp, de); 1222 1223 if (tmpfs_use_nc(fvp)) { 1224 cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp); 1225 } 1226 1227 error = 0; 1228 1229 out_locked: 1230 if (fdvp != tdvp && fdvp != tvp) 1231 VOP_UNLOCK(fdvp); 1232 1233 out: 1234 if (want_seqc_end) { 1235 if (tvp != NULL) 1236 vn_seqc_write_end(tvp); 1237 vn_seqc_write_end(tdvp); 1238 vn_seqc_write_end(fvp); 1239 vn_seqc_write_end(fdvp); 1240 } 1241 1242 /* 1243 * Release target nodes. 1244 * XXX: I don't understand when tdvp can be the same as tvp, but 1245 * other code takes care of this... 1246 */ 1247 if (tdvp == tvp) 1248 vrele(tdvp); 1249 else 1250 vput(tdvp); 1251 if (tvp != NULL) 1252 vput(tvp); 1253 1254 /* Release source nodes. */ 1255 vrele(fdvp); 1256 vrele(fvp); 1257 1258 return (error); 1259 } 1260 1261 static int 1262 tmpfs_mkdir(struct vop_mkdir_args *v) 1263 { 1264 struct vnode *dvp = v->a_dvp; 1265 struct vnode **vpp = v->a_vpp; 1266 struct componentname *cnp = v->a_cnp; 1267 struct vattr *vap = v->a_vap; 1268 1269 MPASS(vap->va_type == VDIR); 1270 1271 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL)); 1272 } 1273 1274 static int 1275 tmpfs_rmdir(struct vop_rmdir_args *v) 1276 { 1277 struct vnode *dvp = v->a_dvp; 1278 struct vnode *vp = v->a_vp; 1279 1280 int error; 1281 struct tmpfs_dirent *de; 1282 struct tmpfs_mount *tmp; 1283 struct tmpfs_node *dnode; 1284 struct tmpfs_node *node; 1285 1286 MPASS(VOP_ISLOCKED(dvp)); 1287 MPASS(VOP_ISLOCKED(vp)); 1288 1289 tmp = VFS_TO_TMPFS(dvp->v_mount); 1290 dnode = VP_TO_TMPFS_DIR(dvp); 1291 node = VP_TO_TMPFS_DIR(vp); 1292 1293 /* Directories with more than two entries ('.' and '..') cannot be 1294 * removed. */ 1295 if (node->tn_size > 0) { 1296 error = ENOTEMPTY; 1297 goto out; 1298 } 1299 1300 if ((dnode->tn_flags & APPEND) 1301 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1302 error = EPERM; 1303 goto out; 1304 } 1305 1306 /* This invariant holds only if we are not trying to remove "..". 1307 * We checked for that above so this is safe now. */ 1308 MPASS(node->tn_dir.tn_parent == dnode); 1309 1310 /* Get the directory entry associated with node (vp). This was 1311 * filled by tmpfs_lookup while looking up the entry. */ 1312 de = tmpfs_dir_lookup(dnode, node, v->a_cnp); 1313 MPASS(TMPFS_DIRENT_MATCHES(de, 1314 v->a_cnp->cn_nameptr, 1315 v->a_cnp->cn_namelen)); 1316 1317 /* Check flags to see if we are allowed to remove the directory. */ 1318 if ((dnode->tn_flags & APPEND) != 0 || 1319 (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) { 1320 error = EPERM; 1321 goto out; 1322 } 1323 1324 /* Detach the directory entry from the directory (dnode). */ 1325 tmpfs_dir_detach(dvp, de); 1326 if (v->a_cnp->cn_flags & DOWHITEOUT) 1327 tmpfs_dir_whiteout_add(dvp, v->a_cnp); 1328 1329 /* No vnode should be allocated for this entry from this point */ 1330 TMPFS_NODE_LOCK(node); 1331 node->tn_links--; 1332 node->tn_dir.tn_parent = NULL; 1333 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1334 node->tn_accessed = true; 1335 1336 TMPFS_NODE_UNLOCK(node); 1337 1338 TMPFS_NODE_LOCK(dnode); 1339 dnode->tn_links--; 1340 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1341 dnode->tn_accessed = true; 1342 TMPFS_NODE_UNLOCK(dnode); 1343 1344 if (tmpfs_use_nc(dvp)) { 1345 cache_vop_rmdir(dvp, vp); 1346 } 1347 1348 /* Free the directory entry we just deleted. Note that the node 1349 * referred by it will not be removed until the vnode is really 1350 * reclaimed. */ 1351 tmpfs_free_dirent(tmp, de); 1352 1353 /* Release the deleted vnode (will destroy the node, notify 1354 * interested parties and clean it from the cache). */ 1355 1356 dnode->tn_status |= TMPFS_NODE_CHANGED; 1357 tmpfs_update(dvp); 1358 1359 error = 0; 1360 1361 out: 1362 return (error); 1363 } 1364 1365 static int 1366 tmpfs_symlink(struct vop_symlink_args *v) 1367 { 1368 struct vnode *dvp = v->a_dvp; 1369 struct vnode **vpp = v->a_vpp; 1370 struct componentname *cnp = v->a_cnp; 1371 struct vattr *vap = v->a_vap; 1372 const char *target = v->a_target; 1373 1374 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */ 1375 MPASS(vap->va_type == VLNK); 1376 #else 1377 vap->va_type = VLNK; 1378 #endif 1379 1380 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, target)); 1381 } 1382 1383 static int 1384 tmpfs_readdir(struct vop_readdir_args *va) 1385 { 1386 struct vnode *vp; 1387 struct uio *uio; 1388 struct tmpfs_mount *tm; 1389 struct tmpfs_node *node; 1390 uint64_t **cookies; 1391 int *eofflag, *ncookies; 1392 ssize_t startresid; 1393 int error, maxcookies; 1394 1395 vp = va->a_vp; 1396 uio = va->a_uio; 1397 eofflag = va->a_eofflag; 1398 cookies = va->a_cookies; 1399 ncookies = va->a_ncookies; 1400 1401 /* This operation only makes sense on directory nodes. */ 1402 if (vp->v_type != VDIR) 1403 return (ENOTDIR); 1404 1405 maxcookies = 0; 1406 node = VP_TO_TMPFS_DIR(vp); 1407 tm = VFS_TO_TMPFS(vp->v_mount); 1408 1409 startresid = uio->uio_resid; 1410 1411 /* Allocate cookies for NFS and compat modules. */ 1412 if (cookies != NULL && ncookies != NULL) { 1413 maxcookies = howmany(node->tn_size, 1414 sizeof(struct tmpfs_dirent)) + 2; 1415 *cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP, 1416 M_WAITOK); 1417 *ncookies = 0; 1418 } 1419 1420 if (cookies == NULL) 1421 error = tmpfs_dir_getdents(tm, node, uio, 0, NULL, NULL); 1422 else 1423 error = tmpfs_dir_getdents(tm, node, uio, maxcookies, *cookies, 1424 ncookies); 1425 1426 /* Buffer was filled without hitting EOF. */ 1427 if (error == EJUSTRETURN) 1428 error = (uio->uio_resid != startresid) ? 0 : EINVAL; 1429 1430 if (error != 0 && cookies != NULL && ncookies != NULL) { 1431 free(*cookies, M_TEMP); 1432 *cookies = NULL; 1433 *ncookies = 0; 1434 } 1435 1436 if (eofflag != NULL) 1437 *eofflag = 1438 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1439 1440 return (error); 1441 } 1442 1443 static int 1444 tmpfs_readlink(struct vop_readlink_args *v) 1445 { 1446 struct vnode *vp = v->a_vp; 1447 struct uio *uio = v->a_uio; 1448 1449 int error; 1450 struct tmpfs_node *node; 1451 1452 MPASS(uio->uio_offset == 0); 1453 MPASS(vp->v_type == VLNK); 1454 1455 node = VP_TO_TMPFS_NODE(vp); 1456 1457 error = uiomove(node->tn_link_target, MIN(node->tn_size, uio->uio_resid), 1458 uio); 1459 tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node); 1460 1461 return (error); 1462 } 1463 1464 /* 1465 * VOP_FPLOOKUP_SYMLINK routines are subject to special circumstances, see 1466 * the comment above cache_fplookup for details. 1467 * 1468 * Check tmpfs_alloc_node for tmpfs-specific synchronisation notes. 1469 */ 1470 static int 1471 tmpfs_fplookup_symlink(struct vop_fplookup_symlink_args *v) 1472 { 1473 struct vnode *vp; 1474 struct tmpfs_node *node; 1475 char *symlink; 1476 1477 vp = v->a_vp; 1478 node = VP_TO_TMPFS_NODE_SMR(vp); 1479 if (__predict_false(node == NULL)) 1480 return (EAGAIN); 1481 if (!atomic_load_char(&node->tn_link_smr)) 1482 return (EAGAIN); 1483 symlink = atomic_load_ptr(&node->tn_link_target); 1484 if (symlink == NULL) 1485 return (EAGAIN); 1486 1487 return (cache_symlink_resolve(v->a_fpl, symlink, node->tn_size)); 1488 } 1489 1490 static int 1491 tmpfs_inactive(struct vop_inactive_args *v) 1492 { 1493 struct vnode *vp; 1494 struct tmpfs_node *node; 1495 1496 vp = v->a_vp; 1497 node = VP_TO_TMPFS_NODE(vp); 1498 if (node->tn_links == 0) 1499 vrecycle(vp); 1500 else 1501 tmpfs_check_mtime(vp); 1502 return (0); 1503 } 1504 1505 static int 1506 tmpfs_need_inactive(struct vop_need_inactive_args *ap) 1507 { 1508 struct vnode *vp; 1509 struct tmpfs_node *node; 1510 struct vm_object *obj; 1511 1512 vp = ap->a_vp; 1513 node = VP_TO_TMPFS_NODE(vp); 1514 if (node->tn_links == 0) 1515 goto need; 1516 if (vp->v_type == VREG) { 1517 obj = vp->v_object; 1518 if (obj->generation != obj->cleangeneration) 1519 goto need; 1520 } 1521 return (0); 1522 need: 1523 return (1); 1524 } 1525 1526 int 1527 tmpfs_reclaim(struct vop_reclaim_args *v) 1528 { 1529 struct vnode *vp; 1530 struct tmpfs_mount *tmp; 1531 struct tmpfs_node *node; 1532 bool unlock; 1533 1534 vp = v->a_vp; 1535 node = VP_TO_TMPFS_NODE(vp); 1536 tmp = VFS_TO_TMPFS(vp->v_mount); 1537 1538 if (vp->v_type == VREG) 1539 tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj); 1540 vp->v_object = NULL; 1541 1542 TMPFS_LOCK(tmp); 1543 TMPFS_NODE_LOCK(node); 1544 tmpfs_free_vp(vp); 1545 1546 /* 1547 * If the node referenced by this vnode was deleted by the user, 1548 * we must free its associated data structures (now that the vnode 1549 * is being reclaimed). 1550 */ 1551 unlock = true; 1552 if (node->tn_links == 0 && 1553 (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) { 1554 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1555 unlock = !tmpfs_free_node_locked(tmp, node, true); 1556 } 1557 1558 if (unlock) { 1559 TMPFS_NODE_UNLOCK(node); 1560 TMPFS_UNLOCK(tmp); 1561 } 1562 1563 MPASS(vp->v_data == NULL); 1564 return (0); 1565 } 1566 1567 int 1568 tmpfs_print(struct vop_print_args *v) 1569 { 1570 struct vnode *vp = v->a_vp; 1571 1572 struct tmpfs_node *node; 1573 1574 node = VP_TO_TMPFS_NODE(vp); 1575 1576 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n", 1577 node, node->tn_flags, (uintmax_t)node->tn_links); 1578 printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n", 1579 node->tn_mode, node->tn_uid, node->tn_gid, 1580 (intmax_t)node->tn_size, node->tn_status); 1581 1582 if (vp->v_type == VFIFO) 1583 fifo_printinfo(vp); 1584 1585 printf("\n"); 1586 1587 return (0); 1588 } 1589 1590 int 1591 tmpfs_pathconf(struct vop_pathconf_args *v) 1592 { 1593 struct vnode *vp = v->a_vp; 1594 int name = v->a_name; 1595 long *retval = v->a_retval; 1596 1597 int error; 1598 1599 error = 0; 1600 1601 switch (name) { 1602 case _PC_LINK_MAX: 1603 *retval = TMPFS_LINK_MAX; 1604 break; 1605 1606 case _PC_SYMLINK_MAX: 1607 *retval = MAXPATHLEN; 1608 break; 1609 1610 case _PC_NAME_MAX: 1611 *retval = NAME_MAX; 1612 break; 1613 1614 case _PC_PIPE_BUF: 1615 if (vp->v_type == VDIR || vp->v_type == VFIFO) 1616 *retval = PIPE_BUF; 1617 else 1618 error = EINVAL; 1619 break; 1620 1621 case _PC_CHOWN_RESTRICTED: 1622 *retval = 1; 1623 break; 1624 1625 case _PC_NO_TRUNC: 1626 *retval = 1; 1627 break; 1628 1629 case _PC_SYNC_IO: 1630 *retval = 1; 1631 break; 1632 1633 case _PC_FILESIZEBITS: 1634 *retval = 64; 1635 break; 1636 1637 case _PC_MIN_HOLE_SIZE: 1638 *retval = PAGE_SIZE; 1639 break; 1640 1641 default: 1642 error = vop_stdpathconf(v); 1643 } 1644 1645 return (error); 1646 } 1647 1648 static int 1649 tmpfs_vptofh(struct vop_vptofh_args *ap) 1650 /* 1651 vop_vptofh { 1652 IN struct vnode *a_vp; 1653 IN struct fid *a_fhp; 1654 }; 1655 */ 1656 { 1657 struct tmpfs_fid_data tfd; 1658 struct tmpfs_node *node; 1659 struct fid *fhp; 1660 1661 node = VP_TO_TMPFS_NODE(ap->a_vp); 1662 fhp = ap->a_fhp; 1663 fhp->fid_len = sizeof(tfd); 1664 1665 /* 1666 * Copy into fid_data from the stack to avoid unaligned pointer use. 1667 * See the comment in sys/mount.h on struct fid for details. 1668 */ 1669 tfd.tfd_id = node->tn_id; 1670 tfd.tfd_gen = node->tn_gen; 1671 memcpy(fhp->fid_data, &tfd, fhp->fid_len); 1672 1673 return (0); 1674 } 1675 1676 static int 1677 tmpfs_whiteout(struct vop_whiteout_args *ap) 1678 { 1679 struct vnode *dvp = ap->a_dvp; 1680 struct componentname *cnp = ap->a_cnp; 1681 struct tmpfs_dirent *de; 1682 1683 switch (ap->a_flags) { 1684 case LOOKUP: 1685 return (0); 1686 case CREATE: 1687 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1688 if (de != NULL) 1689 return (de->td_node == NULL ? 0 : EEXIST); 1690 return (tmpfs_dir_whiteout_add(dvp, cnp)); 1691 case DELETE: 1692 tmpfs_dir_whiteout_remove(dvp, cnp); 1693 return (0); 1694 default: 1695 panic("tmpfs_whiteout: unknown op"); 1696 } 1697 } 1698 1699 static int 1700 tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp, 1701 struct tmpfs_dirent **pde) 1702 { 1703 struct tmpfs_dir_cursor dc; 1704 struct tmpfs_dirent *de; 1705 1706 for (de = tmpfs_dir_first(tnp, &dc); de != NULL; 1707 de = tmpfs_dir_next(tnp, &dc)) { 1708 if (de->td_node == tn) { 1709 *pde = de; 1710 return (0); 1711 } 1712 } 1713 return (ENOENT); 1714 } 1715 1716 static int 1717 tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn, 1718 struct tmpfs_node *tnp, char *buf, size_t *buflen, struct vnode **dvp) 1719 { 1720 struct tmpfs_dirent *de; 1721 int error, i; 1722 1723 error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED, 1724 dvp); 1725 if (error != 0) 1726 return (error); 1727 error = tmpfs_vptocnp_dir(tn, tnp, &de); 1728 if (error == 0) { 1729 i = *buflen; 1730 i -= de->td_namelen; 1731 if (i < 0) { 1732 error = ENOMEM; 1733 } else { 1734 bcopy(de->ud.td_name, buf + i, de->td_namelen); 1735 *buflen = i; 1736 } 1737 } 1738 if (error == 0) { 1739 if (vp != *dvp) 1740 VOP_UNLOCK(*dvp); 1741 } else { 1742 if (vp != *dvp) 1743 vput(*dvp); 1744 else 1745 vrele(vp); 1746 } 1747 return (error); 1748 } 1749 1750 static int 1751 tmpfs_vptocnp(struct vop_vptocnp_args *ap) 1752 { 1753 struct vnode *vp, **dvp; 1754 struct tmpfs_node *tn, *tnp, *tnp1; 1755 struct tmpfs_dirent *de; 1756 struct tmpfs_mount *tm; 1757 char *buf; 1758 size_t *buflen; 1759 int error; 1760 1761 vp = ap->a_vp; 1762 dvp = ap->a_vpp; 1763 buf = ap->a_buf; 1764 buflen = ap->a_buflen; 1765 1766 tm = VFS_TO_TMPFS(vp->v_mount); 1767 tn = VP_TO_TMPFS_NODE(vp); 1768 if (tn->tn_type == VDIR) { 1769 tnp = tn->tn_dir.tn_parent; 1770 if (tnp == NULL) 1771 return (ENOENT); 1772 tmpfs_ref_node(tnp); 1773 error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf, 1774 buflen, dvp); 1775 tmpfs_free_node(tm, tnp); 1776 return (error); 1777 } 1778 restart: 1779 TMPFS_LOCK(tm); 1780 restart_locked: 1781 LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) { 1782 if (tnp->tn_type != VDIR) 1783 continue; 1784 TMPFS_NODE_LOCK(tnp); 1785 tmpfs_ref_node(tnp); 1786 1787 /* 1788 * tn_vnode cannot be instantiated while we hold the 1789 * node lock, so the directory cannot be changed while 1790 * we iterate over it. Do this to avoid instantiating 1791 * vnode for directories which cannot point to our 1792 * node. 1793 */ 1794 error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp, 1795 &de) : 0; 1796 1797 if (error == 0) { 1798 TMPFS_NODE_UNLOCK(tnp); 1799 TMPFS_UNLOCK(tm); 1800 error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen, 1801 dvp); 1802 if (error == 0) { 1803 tmpfs_free_node(tm, tnp); 1804 return (0); 1805 } 1806 if (VN_IS_DOOMED(vp)) { 1807 tmpfs_free_node(tm, tnp); 1808 return (ENOENT); 1809 } 1810 TMPFS_LOCK(tm); 1811 TMPFS_NODE_LOCK(tnp); 1812 } 1813 if (tmpfs_free_node_locked(tm, tnp, false)) { 1814 goto restart; 1815 } else { 1816 KASSERT(tnp->tn_refcount > 0, 1817 ("node %p refcount zero", tnp)); 1818 if (tnp->tn_attached) { 1819 tnp1 = LIST_NEXT(tnp, tn_entries); 1820 TMPFS_NODE_UNLOCK(tnp); 1821 } else { 1822 TMPFS_NODE_UNLOCK(tnp); 1823 goto restart_locked; 1824 } 1825 } 1826 } 1827 TMPFS_UNLOCK(tm); 1828 return (ENOENT); 1829 } 1830 1831 static off_t 1832 tmpfs_seek_data_locked(vm_object_t obj, off_t noff) 1833 { 1834 vm_page_t m; 1835 vm_pindex_t p, p_m, p_swp; 1836 1837 p = OFF_TO_IDX(noff); 1838 m = vm_page_find_least(obj, p); 1839 1840 /* 1841 * Microoptimize the most common case for SEEK_DATA, where 1842 * there is no hole and the page is resident. 1843 */ 1844 if (m != NULL && vm_page_any_valid(m) && m->pindex == p) 1845 return (noff); 1846 1847 p_swp = swap_pager_find_least(obj, p); 1848 if (p_swp == p) 1849 return (noff); 1850 1851 p_m = m == NULL ? obj->size : m->pindex; 1852 return (IDX_TO_OFF(MIN(p_m, p_swp))); 1853 } 1854 1855 static off_t 1856 tmpfs_seek_next(off_t noff) 1857 { 1858 return (noff + PAGE_SIZE - (noff & PAGE_MASK)); 1859 } 1860 1861 static int 1862 tmpfs_seek_clamp(struct tmpfs_node *tn, off_t *noff, bool seekdata) 1863 { 1864 if (*noff < tn->tn_size) 1865 return (0); 1866 if (seekdata) 1867 return (ENXIO); 1868 *noff = tn->tn_size; 1869 return (0); 1870 } 1871 1872 static off_t 1873 tmpfs_seek_hole_locked(vm_object_t obj, off_t noff) 1874 { 1875 vm_page_t m; 1876 vm_pindex_t p, p_swp; 1877 1878 for (;; noff = tmpfs_seek_next(noff)) { 1879 /* 1880 * Walk over the largest sequential run of the valid pages. 1881 */ 1882 for (m = vm_page_lookup(obj, OFF_TO_IDX(noff)); 1883 m != NULL && vm_page_any_valid(m); 1884 m = vm_page_next(m), noff = tmpfs_seek_next(noff)) 1885 ; 1886 1887 /* 1888 * Found a hole in the object's page queue. Check if 1889 * there is a hole in the swap at the same place. 1890 */ 1891 p = OFF_TO_IDX(noff); 1892 p_swp = swap_pager_find_least(obj, p); 1893 if (p_swp != p) { 1894 noff = IDX_TO_OFF(p); 1895 break; 1896 } 1897 } 1898 return (noff); 1899 } 1900 1901 static int 1902 tmpfs_seek_datahole(struct vnode *vp, off_t *off, bool seekdata) 1903 { 1904 struct tmpfs_node *tn; 1905 vm_object_t obj; 1906 off_t noff; 1907 int error; 1908 1909 if (vp->v_type != VREG) 1910 return (ENOTTY); 1911 tn = VP_TO_TMPFS_NODE(vp); 1912 noff = *off; 1913 if (noff < 0) 1914 return (ENXIO); 1915 error = tmpfs_seek_clamp(tn, &noff, seekdata); 1916 if (error != 0) 1917 return (error); 1918 obj = tn->tn_reg.tn_aobj; 1919 1920 VM_OBJECT_RLOCK(obj); 1921 noff = seekdata ? tmpfs_seek_data_locked(obj, noff) : 1922 tmpfs_seek_hole_locked(obj, noff); 1923 VM_OBJECT_RUNLOCK(obj); 1924 1925 error = tmpfs_seek_clamp(tn, &noff, seekdata); 1926 if (error == 0) 1927 *off = noff; 1928 return (error); 1929 } 1930 1931 static int 1932 tmpfs_ioctl(struct vop_ioctl_args *ap) 1933 { 1934 struct vnode *vp = ap->a_vp; 1935 int error = 0; 1936 1937 switch (ap->a_command) { 1938 case FIOSEEKDATA: 1939 case FIOSEEKHOLE: 1940 error = vn_lock(vp, LK_SHARED); 1941 if (error != 0) { 1942 error = EBADF; 1943 break; 1944 } 1945 error = tmpfs_seek_datahole(vp, (off_t *)ap->a_data, 1946 ap->a_command == FIOSEEKDATA); 1947 VOP_UNLOCK(vp); 1948 break; 1949 default: 1950 error = ENOTTY; 1951 break; 1952 } 1953 return (error); 1954 } 1955 1956 /* 1957 * Vnode operations vector used for files stored in a tmpfs file system. 1958 */ 1959 struct vop_vector tmpfs_vnodeop_entries = { 1960 .vop_default = &default_vnodeops, 1961 .vop_lookup = vfs_cache_lookup, 1962 .vop_cachedlookup = tmpfs_cached_lookup, 1963 .vop_create = tmpfs_create, 1964 .vop_mknod = tmpfs_mknod, 1965 .vop_open = tmpfs_open, 1966 .vop_close = tmpfs_close, 1967 .vop_fplookup_vexec = tmpfs_fplookup_vexec, 1968 .vop_fplookup_symlink = tmpfs_fplookup_symlink, 1969 .vop_access = tmpfs_access, 1970 .vop_stat = tmpfs_stat, 1971 .vop_getattr = tmpfs_getattr, 1972 .vop_setattr = tmpfs_setattr, 1973 .vop_read = tmpfs_read, 1974 .vop_read_pgcache = tmpfs_read_pgcache, 1975 .vop_write = tmpfs_write, 1976 .vop_deallocate = tmpfs_deallocate, 1977 .vop_fsync = tmpfs_fsync, 1978 .vop_remove = tmpfs_remove, 1979 .vop_link = tmpfs_link, 1980 .vop_rename = tmpfs_rename, 1981 .vop_mkdir = tmpfs_mkdir, 1982 .vop_rmdir = tmpfs_rmdir, 1983 .vop_symlink = tmpfs_symlink, 1984 .vop_readdir = tmpfs_readdir, 1985 .vop_readlink = tmpfs_readlink, 1986 .vop_inactive = tmpfs_inactive, 1987 .vop_need_inactive = tmpfs_need_inactive, 1988 .vop_reclaim = tmpfs_reclaim, 1989 .vop_print = tmpfs_print, 1990 .vop_pathconf = tmpfs_pathconf, 1991 .vop_vptofh = tmpfs_vptofh, 1992 .vop_whiteout = tmpfs_whiteout, 1993 .vop_bmap = VOP_EOPNOTSUPP, 1994 .vop_vptocnp = tmpfs_vptocnp, 1995 .vop_lock1 = vop_lock, 1996 .vop_unlock = vop_unlock, 1997 .vop_islocked = vop_islocked, 1998 .vop_add_writecount = vop_stdadd_writecount_nomsync, 1999 .vop_ioctl = tmpfs_ioctl, 2000 }; 2001 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_entries); 2002 2003 /* 2004 * Same vector for mounts which do not use namecache. 2005 */ 2006 struct vop_vector tmpfs_vnodeop_nonc_entries = { 2007 .vop_default = &tmpfs_vnodeop_entries, 2008 .vop_lookup = tmpfs_lookup, 2009 }; 2010 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_nonc_entries); 2011