| 1 | /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */ |
| 2 | |
| 3 | /*- |
| 4 | * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc. |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * This code is derived from software contributed to The NetBSD Foundation |
| 8 | * by Julio M. Merino Vidal, developed as part of Google's Summer of Code |
| 9 | * 2005 program. |
| 10 | * |
| 11 | * Redistribution and use in source and binary forms, with or without |
| 12 | * modification, are permitted provided that the following conditions |
| 13 | * are met: |
| 14 | * 1. Redistributions of source code must retain the above copyright |
| 15 | * notice, this list of conditions and the following disclaimer. |
| 16 | * 2. Redistributions in binary form must reproduce the above copyright |
| 17 | * notice, this list of conditions and the following disclaimer in the |
| 18 | * documentation and/or other materials provided with the distribution. |
| 19 | * |
| 20 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS |
| 21 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
| 22 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| 23 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS |
| 24 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 25 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 26 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 27 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 28 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 29 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
| 30 | * POSSIBILITY OF SUCH DAMAGE. |
| 31 | */ |
| 32 | |
| 33 | /* |
| 34 | * tmpfs vnode interface. |
| 35 | */ |
| 36 | #include <sys/cdefs.h> |
| 37 | |
| 38 | #include <sys/kernel.h> |
| 39 | #include <sys/kern_syscall.h> |
| 40 | #include <sys/param.h> |
| 41 | #include <sys/fcntl.h> |
| 42 | #include <sys/lockf.h> |
| 43 | #include <sys/priv.h> |
| 44 | #include <sys/proc.h> |
| 45 | #include <sys/resourcevar.h> |
| 46 | #include <sys/sched.h> |
| 47 | #include <sys/stat.h> |
| 48 | #include <sys/systm.h> |
| 49 | #include <sys/unistd.h> |
| 50 | #include <sys/vfsops.h> |
| 51 | #include <sys/vnode.h> |
| 52 | |
| 53 | #include <sys/mplock2.h> |
| 54 | |
| 55 | #include <vm/vm.h> |
| 56 | #include <vm/vm_object.h> |
| 57 | #include <vm/vm_page.h> |
| 58 | #include <vm/vm_pager.h> |
| 59 | |
| 60 | #include <vfs/fifofs/fifo.h> |
| 61 | #include <vfs/tmpfs/tmpfs_vnops.h> |
| 62 | #include <vfs/tmpfs/tmpfs.h> |
| 63 | |
| 64 | MALLOC_DECLARE(M_TMPFS); |
| 65 | |
| 66 | /* --------------------------------------------------------------------- */ |
| 67 | |
| 68 | static int |
| 69 | tmpfs_nresolve(struct vop_nresolve_args *v) |
| 70 | { |
| 71 | struct vnode *dvp = v->a_dvp; |
| 72 | struct vnode *vp = NULL; |
| 73 | struct namecache *ncp = v->a_nch->ncp; |
| 74 | struct tmpfs_node *tnode; |
| 75 | |
| 76 | int error; |
| 77 | struct tmpfs_dirent *de; |
| 78 | struct tmpfs_node *dnode; |
| 79 | |
| 80 | dnode = VP_TO_TMPFS_DIR(dvp); |
| 81 | |
| 82 | de = tmpfs_dir_lookup(dnode, NULL, ncp); |
| 83 | if (de == NULL) { |
| 84 | error = ENOENT; |
| 85 | } else { |
| 86 | /* |
| 87 | * Allocate a vnode for the node we found. |
| 88 | */ |
| 89 | tnode = de->td_node; |
| 90 | error = tmpfs_alloc_vp(dvp->v_mount, tnode, |
| 91 | LK_EXCLUSIVE | LK_RETRY, &vp); |
| 92 | if (error) |
| 93 | goto out; |
| 94 | KKASSERT(vp); |
| 95 | } |
| 96 | |
| 97 | out: |
| 98 | /* |
| 99 | * Store the result of this lookup in the cache. Avoid this if the |
| 100 | * request was for creation, as it does not improve timings on |
| 101 | * emprical tests. |
| 102 | */ |
| 103 | if (vp) { |
| 104 | vn_unlock(vp); |
| 105 | cache_setvp(v->a_nch, vp); |
| 106 | vrele(vp); |
| 107 | } else if (error == ENOENT) { |
| 108 | cache_setvp(v->a_nch, NULL); |
| 109 | } |
| 110 | return error; |
| 111 | } |
| 112 | |
| 113 | static int |
| 114 | tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v) |
| 115 | { |
| 116 | struct vnode *dvp = v->a_dvp; |
| 117 | struct vnode **vpp = v->a_vpp; |
| 118 | struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp); |
| 119 | struct ucred *cred = v->a_cred; |
| 120 | int error; |
| 121 | |
| 122 | *vpp = NULL; |
| 123 | /* Check accessibility of requested node as a first step. */ |
| 124 | error = VOP_ACCESS(dvp, VEXEC, cred); |
| 125 | if (error != 0) |
| 126 | return error; |
| 127 | |
| 128 | if (dnode->tn_dir.tn_parent != NULL) { |
| 129 | /* Allocate a new vnode on the matching entry. */ |
| 130 | error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent, |
| 131 | LK_EXCLUSIVE | LK_RETRY, vpp); |
| 132 | |
| 133 | if (*vpp) |
| 134 | vn_unlock(*vpp); |
| 135 | } |
| 136 | |
| 137 | return (*vpp == NULL) ? ENOENT : 0; |
| 138 | } |
| 139 | |
| 140 | /* --------------------------------------------------------------------- */ |
| 141 | |
| 142 | static int |
| 143 | tmpfs_ncreate(struct vop_ncreate_args *v) |
| 144 | { |
| 145 | struct vnode *dvp = v->a_dvp; |
| 146 | struct vnode **vpp = v->a_vpp; |
| 147 | struct namecache *ncp = v->a_nch->ncp; |
| 148 | struct vattr *vap = v->a_vap; |
| 149 | struct ucred *cred = v->a_cred; |
| 150 | int error; |
| 151 | |
| 152 | KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK); |
| 153 | |
| 154 | error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); |
| 155 | if (error == 0) { |
| 156 | cache_setunresolved(v->a_nch); |
| 157 | cache_setvp(v->a_nch, *vpp); |
| 158 | } |
| 159 | |
| 160 | return error; |
| 161 | } |
| 162 | /* --------------------------------------------------------------------- */ |
| 163 | |
| 164 | static int |
| 165 | tmpfs_nmknod(struct vop_nmknod_args *v) |
| 166 | { |
| 167 | struct vnode *dvp = v->a_dvp; |
| 168 | struct vnode **vpp = v->a_vpp; |
| 169 | struct namecache *ncp = v->a_nch->ncp; |
| 170 | struct vattr *vap = v->a_vap; |
| 171 | struct ucred *cred = v->a_cred; |
| 172 | int error; |
| 173 | |
| 174 | if (vap->va_type != VBLK && vap->va_type != VCHR && |
| 175 | vap->va_type != VFIFO) |
| 176 | return EINVAL; |
| 177 | |
| 178 | error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); |
| 179 | if (error == 0) { |
| 180 | cache_setunresolved(v->a_nch); |
| 181 | cache_setvp(v->a_nch, *vpp); |
| 182 | } |
| 183 | |
| 184 | return error; |
| 185 | } |
| 186 | |
| 187 | /* --------------------------------------------------------------------- */ |
| 188 | |
| 189 | static int |
| 190 | tmpfs_open(struct vop_open_args *v) |
| 191 | { |
| 192 | struct vnode *vp = v->a_vp; |
| 193 | int mode = v->a_mode; |
| 194 | |
| 195 | int error; |
| 196 | struct tmpfs_node *node; |
| 197 | |
| 198 | node = VP_TO_TMPFS_NODE(vp); |
| 199 | |
| 200 | /* The file is still active but all its names have been removed |
| 201 | * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as |
| 202 | * it is about to die. */ |
| 203 | if (node->tn_links < 1) |
| 204 | return (ENOENT); |
| 205 | |
| 206 | /* If the file is marked append-only, deny write requests. */ |
| 207 | if ((node->tn_flags & APPEND) && |
| 208 | (mode & (FWRITE | O_APPEND)) == FWRITE) { |
| 209 | error = EPERM; |
| 210 | } else { |
| 211 | return (vop_stdopen(v)); |
| 212 | } |
| 213 | return error; |
| 214 | } |
| 215 | |
| 216 | /* --------------------------------------------------------------------- */ |
| 217 | |
| 218 | static int |
| 219 | tmpfs_close(struct vop_close_args *v) |
| 220 | { |
| 221 | struct vnode *vp = v->a_vp; |
| 222 | struct tmpfs_node *node; |
| 223 | |
| 224 | node = VP_TO_TMPFS_NODE(vp); |
| 225 | |
| 226 | if (node->tn_links > 0) { |
| 227 | /* Update node times. No need to do it if the node has |
| 228 | * been deleted, because it will vanish after we return. */ |
| 229 | tmpfs_update(vp); |
| 230 | } |
| 231 | |
| 232 | return vop_stdclose(v); |
| 233 | } |
| 234 | |
| 235 | /* --------------------------------------------------------------------- */ |
| 236 | |
| 237 | int |
| 238 | tmpfs_access(struct vop_access_args *v) |
| 239 | { |
| 240 | struct vnode *vp = v->a_vp; |
| 241 | int error; |
| 242 | struct tmpfs_node *node; |
| 243 | |
| 244 | node = VP_TO_TMPFS_NODE(vp); |
| 245 | |
| 246 | switch (vp->v_type) { |
| 247 | case VDIR: |
| 248 | /* FALLTHROUGH */ |
| 249 | case VLNK: |
| 250 | /* FALLTHROUGH */ |
| 251 | case VREG: |
| 252 | if (VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) { |
| 253 | error = EROFS; |
| 254 | goto out; |
| 255 | } |
| 256 | break; |
| 257 | |
| 258 | case VBLK: |
| 259 | /* FALLTHROUGH */ |
| 260 | case VCHR: |
| 261 | /* FALLTHROUGH */ |
| 262 | case VSOCK: |
| 263 | /* FALLTHROUGH */ |
| 264 | case VFIFO: |
| 265 | break; |
| 266 | |
| 267 | default: |
| 268 | error = EINVAL; |
| 269 | goto out; |
| 270 | } |
| 271 | |
| 272 | if (VWRITE && node->tn_flags & IMMUTABLE) { |
| 273 | error = EPERM; |
| 274 | goto out; |
| 275 | } |
| 276 | |
| 277 | error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0); |
| 278 | |
| 279 | out: |
| 280 | |
| 281 | return error; |
| 282 | } |
| 283 | |
| 284 | /* --------------------------------------------------------------------- */ |
| 285 | |
| 286 | int |
| 287 | tmpfs_getattr(struct vop_getattr_args *v) |
| 288 | { |
| 289 | struct vnode *vp = v->a_vp; |
| 290 | struct vattr *vap = v->a_vap; |
| 291 | struct tmpfs_node *node; |
| 292 | |
| 293 | node = VP_TO_TMPFS_NODE(vp); |
| 294 | |
| 295 | tmpfs_update(vp); |
| 296 | |
| 297 | vap->va_type = vp->v_type; |
| 298 | vap->va_mode = node->tn_mode; |
| 299 | vap->va_nlink = node->tn_links; |
| 300 | vap->va_uid = node->tn_uid; |
| 301 | vap->va_gid = node->tn_gid; |
| 302 | vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; |
| 303 | vap->va_fileid = node->tn_id; |
| 304 | vap->va_size = node->tn_size; |
| 305 | vap->va_blocksize = PAGE_SIZE; |
| 306 | vap->va_atime.tv_sec = node->tn_atime; |
| 307 | vap->va_atime.tv_nsec = node->tn_atimensec; |
| 308 | vap->va_mtime.tv_sec = node->tn_mtime; |
| 309 | vap->va_mtime.tv_nsec = node->tn_mtimensec; |
| 310 | vap->va_ctime.tv_sec = node->tn_ctime; |
| 311 | vap->va_ctime.tv_nsec = node->tn_ctimensec; |
| 312 | vap->va_gen = node->tn_gen; |
| 313 | vap->va_flags = node->tn_flags; |
| 314 | if (vp->v_type == VBLK || vp->v_type == VCHR) |
| 315 | { |
| 316 | vap->va_rmajor = umajor(node->tn_rdev); |
| 317 | vap->va_rminor = uminor(node->tn_rdev); |
| 318 | } |
| 319 | vap->va_bytes = round_page(node->tn_size); |
| 320 | vap->va_filerev = 0; |
| 321 | |
| 322 | return 0; |
| 323 | } |
| 324 | |
| 325 | /* --------------------------------------------------------------------- */ |
| 326 | |
| 327 | int |
| 328 | tmpfs_setattr(struct vop_setattr_args *v) |
| 329 | { |
| 330 | struct vnode *vp = v->a_vp; |
| 331 | struct vattr *vap = v->a_vap; |
| 332 | struct ucred *cred = v->a_cred; |
| 333 | int error = 0; |
| 334 | |
| 335 | if (error == 0 && (vap->va_flags != VNOVAL)) |
| 336 | error = tmpfs_chflags(vp, vap->va_flags, cred); |
| 337 | |
| 338 | if (error == 0 && (vap->va_size != VNOVAL)) |
| 339 | error = tmpfs_chsize(vp, vap->va_size, cred); |
| 340 | |
| 341 | if (error == 0 && (vap->va_uid != (uid_t)VNOVAL || |
| 342 | vap->va_gid != (gid_t)VNOVAL)) { |
| 343 | error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred); |
| 344 | } |
| 345 | |
| 346 | if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) |
| 347 | error = tmpfs_chmod(vp, vap->va_mode, cred); |
| 348 | |
| 349 | if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && |
| 350 | vap->va_atime.tv_nsec != VNOVAL) || |
| 351 | (vap->va_mtime.tv_sec != VNOVAL && |
| 352 | vap->va_mtime.tv_nsec != VNOVAL) )) { |
| 353 | error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime, |
| 354 | vap->va_vaflags, cred); |
| 355 | } |
| 356 | |
| 357 | /* Update the node times. We give preference to the error codes |
| 358 | * generated by this function rather than the ones that may arise |
| 359 | * from tmpfs_update. */ |
| 360 | tmpfs_update(vp); |
| 361 | |
| 362 | return error; |
| 363 | } |
| 364 | |
| 365 | /* --------------------------------------------------------------------- */ |
| 366 | |
| 367 | /* |
| 368 | * fsync is usually a NOP, but we must take action when unmounting or |
| 369 | * when recycling. |
| 370 | */ |
| 371 | static int |
| 372 | tmpfs_fsync(struct vop_fsync_args *v) |
| 373 | { |
| 374 | struct tmpfs_mount *tmp; |
| 375 | struct tmpfs_node *node; |
| 376 | struct vnode *vp = v->a_vp; |
| 377 | |
| 378 | tmp = VFS_TO_TMPFS(vp->v_mount); |
| 379 | node = VP_TO_TMPFS_NODE(vp); |
| 380 | |
| 381 | tmpfs_update(vp); |
| 382 | if (vp->v_type == VREG) { |
| 383 | if (vp->v_flag & VRECLAIMED) { |
| 384 | if (node->tn_links == 0) |
| 385 | tmpfs_truncate(vp, 0); |
| 386 | else |
| 387 | vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL); |
| 388 | } |
| 389 | } |
| 390 | return 0; |
| 391 | } |
| 392 | |
| 393 | /* --------------------------------------------------------------------- */ |
| 394 | |
| 395 | static int |
| 396 | tmpfs_read (struct vop_read_args *ap) |
| 397 | { |
| 398 | struct buf *bp; |
| 399 | struct vnode *vp = ap->a_vp; |
| 400 | struct uio *uio = ap->a_uio; |
| 401 | struct tmpfs_node *node; |
| 402 | off_t base_offset; |
| 403 | size_t offset; |
| 404 | size_t len; |
| 405 | int got_mplock; |
| 406 | int error; |
| 407 | |
| 408 | error = 0; |
| 409 | if (uio->uio_resid == 0) { |
| 410 | return error; |
| 411 | } |
| 412 | |
| 413 | node = VP_TO_TMPFS_NODE(vp); |
| 414 | |
| 415 | if (uio->uio_offset < 0) |
| 416 | return (EINVAL); |
| 417 | if (vp->v_type != VREG) |
| 418 | return (EINVAL); |
| 419 | |
| 420 | #ifdef SMP |
| 421 | if(curthread->td_mpcount) |
| 422 | got_mplock = -1; |
| 423 | else |
| 424 | got_mplock = 0; |
| 425 | #else |
| 426 | got_mplock = -1; |
| 427 | #endif |
| 428 | |
| 429 | while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) { |
| 430 | /* |
| 431 | * Use buffer cache I/O (via tmpfs_strategy) |
| 432 | */ |
| 433 | offset = (size_t)uio->uio_offset & BMASK; |
| 434 | base_offset = (off_t)uio->uio_offset - offset; |
| 435 | bp = getcacheblk(vp, base_offset); |
| 436 | if (bp == NULL) |
| 437 | { |
| 438 | if (got_mplock == 0) { |
| 439 | got_mplock = 1; |
| 440 | get_mplock(); |
| 441 | } |
| 442 | |
| 443 | error = bread(vp, base_offset, BSIZE, &bp); |
| 444 | if (error) { |
| 445 | brelse(bp); |
| 446 | kprintf("tmpfs_read bread error %d\n", error); |
| 447 | break; |
| 448 | } |
| 449 | } |
| 450 | |
| 451 | if (got_mplock == 0) { |
| 452 | got_mplock = 1; |
| 453 | get_mplock(); |
| 454 | } |
| 455 | |
| 456 | /* |
| 457 | * Figure out how many bytes we can actually copy this loop. |
| 458 | */ |
| 459 | len = BSIZE - offset; |
| 460 | if (len > uio->uio_resid) |
| 461 | len = uio->uio_resid; |
| 462 | if (len > node->tn_size - uio->uio_offset) |
| 463 | len = (size_t)(node->tn_size - uio->uio_offset); |
| 464 | |
| 465 | error = uiomove((char *)bp->b_data + offset, len, uio); |
| 466 | bqrelse(bp); |
| 467 | if (error) { |
| 468 | kprintf("tmpfs_read uiomove error %d\n", error); |
| 469 | break; |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | if (got_mplock > 0) |
| 474 | rel_mplock(); |
| 475 | |
| 476 | TMPFS_NODE_LOCK(node); |
| 477 | node->tn_status |= TMPFS_NODE_ACCESSED; |
| 478 | TMPFS_NODE_UNLOCK(node); |
| 479 | |
| 480 | return(error); |
| 481 | } |
| 482 | |
| 483 | static int |
| 484 | tmpfs_write (struct vop_write_args *ap) |
| 485 | { |
| 486 | struct buf *bp; |
| 487 | struct vnode *vp = ap->a_vp; |
| 488 | struct uio *uio = ap->a_uio; |
| 489 | struct thread *td = uio->uio_td; |
| 490 | struct tmpfs_node *node; |
| 491 | boolean_t extended; |
| 492 | off_t oldsize; |
| 493 | int error; |
| 494 | off_t base_offset; |
| 495 | size_t offset; |
| 496 | size_t len; |
| 497 | struct rlimit limit; |
| 498 | int got_mplock; |
| 499 | int trivial = 0; |
| 500 | |
| 501 | error = 0; |
| 502 | if (uio->uio_resid == 0) { |
| 503 | return error; |
| 504 | } |
| 505 | |
| 506 | node = VP_TO_TMPFS_NODE(vp); |
| 507 | |
| 508 | if (vp->v_type != VREG) |
| 509 | return (EINVAL); |
| 510 | |
| 511 | oldsize = node->tn_size; |
| 512 | if (ap->a_ioflag & IO_APPEND) |
| 513 | uio->uio_offset = node->tn_size; |
| 514 | |
| 515 | /* |
| 516 | * Check for illegal write offsets. |
| 517 | */ |
| 518 | if (uio->uio_offset + uio->uio_resid > |
| 519 | VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) |
| 520 | return (EFBIG); |
| 521 | |
| 522 | if (vp->v_type == VREG && td != NULL) { |
| 523 | error = kern_getrlimit(RLIMIT_FSIZE, &limit); |
| 524 | if (error != 0) |
| 525 | return error; |
| 526 | if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) { |
| 527 | ksignal(td->td_proc, SIGXFSZ); |
| 528 | return (EFBIG); |
| 529 | } |
| 530 | } |
| 531 | |
| 532 | |
| 533 | /* |
| 534 | * Extend the file's size if necessary |
| 535 | */ |
| 536 | extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size); |
| 537 | |
| 538 | #ifdef SMP |
| 539 | if (curthread->td_mpcount) { |
| 540 | got_mplock = -1; |
| 541 | } else { |
| 542 | got_mplock = 1; |
| 543 | get_mplock(); |
| 544 | } |
| 545 | #else |
| 546 | got_mplock = -1; |
| 547 | #endif |
| 548 | while (uio->uio_resid > 0) { |
| 549 | /* |
| 550 | * Use buffer cache I/O (via tmpfs_strategy) |
| 551 | */ |
| 552 | offset = (size_t)uio->uio_offset & BMASK; |
| 553 | base_offset = (off_t)uio->uio_offset - offset; |
| 554 | len = BSIZE - offset; |
| 555 | if (len > uio->uio_resid) |
| 556 | len = uio->uio_resid; |
| 557 | |
| 558 | if ((uio->uio_offset + len) > node->tn_size) { |
| 559 | trivial = (uio->uio_offset <= node->tn_size); |
| 560 | error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial); |
| 561 | if (error) |
| 562 | break; |
| 563 | } |
| 564 | |
| 565 | /* |
| 566 | * Read to fill in any gaps. Theoretically we could |
| 567 | * optimize this if the write covers the entire buffer |
| 568 | * and is not a UIO_NOCOPY write, however this can lead |
| 569 | * to a security violation exposing random kernel memory |
| 570 | * (whatever junk was in the backing VM pages before). |
| 571 | * |
| 572 | * So just use bread() to do the right thing. |
| 573 | */ |
| 574 | error = bread(vp, base_offset, BSIZE, &bp); |
| 575 | error = uiomove((char *)bp->b_data + offset, len, uio); |
| 576 | if (error) { |
| 577 | kprintf("tmpfs_write uiomove error %d\n", error); |
| 578 | brelse(bp); |
| 579 | break; |
| 580 | } |
| 581 | |
| 582 | if (uio->uio_offset > node->tn_size) |
| 583 | node->tn_size = uio->uio_offset; |
| 584 | |
| 585 | /* |
| 586 | * The data has been loaded into the buffer, write it out. |
| 587 | * |
| 588 | * We want tmpfs to be able to use all available ram, not |
| 589 | * just the buffer cache, so if not explicitly paging we |
| 590 | * use buwrite() to leave the buffer clean but mark all the |
| 591 | * VM pages valid+dirty. |
| 592 | * |
| 593 | * When the kernel is paging, either via normal pageout |
| 594 | * operation or when cleaning the object during a recycle, |
| 595 | * the underlying VM pages are going to get thrown away |
| 596 | * so we MUST write them to swap. |
| 597 | * |
| 598 | * XXX unfortunately this catches msync() system calls too |
| 599 | * for the moment. |
| 600 | */ |
| 601 | if (ap->a_ioflag & IO_SYNC) { |
| 602 | bwrite(bp); |
| 603 | } else if ((ap->a_ioflag & IO_ASYNC) || |
| 604 | (uio->uio_segflg == UIO_NOCOPY)) { |
| 605 | bawrite(bp); |
| 606 | } else { |
| 607 | buwrite(bp); |
| 608 | } |
| 609 | |
| 610 | if (bp->b_error) { |
| 611 | kprintf("tmpfs_write bwrite error %d\n", error); |
| 612 | break; |
| 613 | } |
| 614 | } |
| 615 | |
| 616 | if (got_mplock > 0) |
| 617 | rel_mplock(); |
| 618 | |
| 619 | if (error) { |
| 620 | if (extended) |
| 621 | (void)tmpfs_reg_resize(vp, oldsize, trivial); |
| 622 | return error; |
| 623 | } |
| 624 | |
| 625 | TMPFS_NODE_LOCK(node); |
| 626 | node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | |
| 627 | (extended? TMPFS_NODE_CHANGED : 0); |
| 628 | |
| 629 | if (node->tn_mode & (S_ISUID | S_ISGID)) { |
| 630 | if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) |
| 631 | node->tn_mode &= ~(S_ISUID | S_ISGID); |
| 632 | } |
| 633 | TMPFS_NODE_UNLOCK(node); |
| 634 | |
| 635 | return(error); |
| 636 | } |
| 637 | |
| 638 | static int |
| 639 | tmpfs_advlock (struct vop_advlock_args *ap) |
| 640 | { |
| 641 | struct tmpfs_node *node; |
| 642 | struct vnode *vp = ap->a_vp; |
| 643 | |
| 644 | node = VP_TO_TMPFS_NODE(vp); |
| 645 | |
| 646 | return (lf_advlock(ap, &node->tn_advlock, node->tn_size)); |
| 647 | } |
| 648 | |
| 649 | |
| 650 | static int |
| 651 | tmpfs_strategy(struct vop_strategy_args *ap) |
| 652 | { |
| 653 | struct bio *bio = ap->a_bio; |
| 654 | struct buf *bp = bio->bio_buf; |
| 655 | struct vnode *vp = ap->a_vp; |
| 656 | struct tmpfs_node *node; |
| 657 | vm_object_t uobj; |
| 658 | |
| 659 | if (vp->v_type != VREG) { |
| 660 | bp->b_resid = bp->b_bcount; |
| 661 | bp->b_flags |= B_ERROR | B_INVAL; |
| 662 | bp->b_error = EINVAL; |
| 663 | biodone(bio); |
| 664 | return(0); |
| 665 | } |
| 666 | |
| 667 | node = VP_TO_TMPFS_NODE(vp); |
| 668 | |
| 669 | uobj = node->tn_reg.tn_aobj; |
| 670 | |
| 671 | /* |
| 672 | * Call swap_pager_strategy to read or write between the VM |
| 673 | * object and the buffer cache. |
| 674 | */ |
| 675 | swap_pager_strategy(uobj, bio); |
| 676 | |
| 677 | return 0; |
| 678 | } |
| 679 | |
| 680 | static int |
| 681 | tmpfs_bmap(struct vop_bmap_args *ap) |
| 682 | { |
| 683 | if (ap->a_doffsetp != NULL) |
| 684 | *ap->a_doffsetp = ap->a_loffset; |
| 685 | if (ap->a_runp != NULL) |
| 686 | *ap->a_runp = 0; |
| 687 | if (ap->a_runb != NULL) |
| 688 | *ap->a_runb = 0; |
| 689 | |
| 690 | return 0; |
| 691 | } |
| 692 | |
| 693 | /* --------------------------------------------------------------------- */ |
| 694 | |
| 695 | static int |
| 696 | tmpfs_nremove(struct vop_nremove_args *v) |
| 697 | { |
| 698 | struct vnode *dvp = v->a_dvp; |
| 699 | struct namecache *ncp = v->a_nch->ncp; |
| 700 | struct vnode *vp; |
| 701 | int error; |
| 702 | struct tmpfs_dirent *de; |
| 703 | struct tmpfs_mount *tmp; |
| 704 | struct tmpfs_node *dnode; |
| 705 | struct tmpfs_node *node; |
| 706 | |
| 707 | /* |
| 708 | * We have to acquire the vp from v->a_nch because |
| 709 | * we will likely unresolve the namecache entry, and |
| 710 | * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim |
| 711 | * sequence to recover space from the file. |
| 712 | */ |
| 713 | error = cache_vref(v->a_nch, v->a_cred, &vp); |
| 714 | KKASSERT(error == 0); |
| 715 | |
| 716 | if (vp->v_type == VDIR) { |
| 717 | error = EISDIR; |
| 718 | goto out; |
| 719 | } |
| 720 | |
| 721 | dnode = VP_TO_TMPFS_DIR(dvp); |
| 722 | node = VP_TO_TMPFS_NODE(vp); |
| 723 | tmp = VFS_TO_TMPFS(vp->v_mount); |
| 724 | de = tmpfs_dir_lookup(dnode, node, ncp); |
| 725 | if (de == NULL) { |
| 726 | error = ENOENT; |
| 727 | goto out; |
| 728 | } |
| 729 | |
| 730 | /* Files marked as immutable or append-only cannot be deleted. */ |
| 731 | if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || |
| 732 | (dnode->tn_flags & APPEND)) { |
| 733 | error = EPERM; |
| 734 | goto out; |
| 735 | } |
| 736 | |
| 737 | /* Remove the entry from the directory; as it is a file, we do not |
| 738 | * have to change the number of hard links of the directory. */ |
| 739 | tmpfs_dir_detach(dnode, de); |
| 740 | |
| 741 | /* Free the directory entry we just deleted. Note that the node |
| 742 | * referred by it will not be removed until the vnode is really |
| 743 | * reclaimed. */ |
| 744 | tmpfs_free_dirent(tmp, de); |
| 745 | |
| 746 | if (node->tn_links > 0) { |
| 747 | TMPFS_NODE_LOCK(node); |
| 748 | node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ |
| 749 | TMPFS_NODE_MODIFIED; |
| 750 | TMPFS_NODE_UNLOCK(node); |
| 751 | } |
| 752 | |
| 753 | cache_setunresolved(v->a_nch); |
| 754 | cache_setvp(v->a_nch, NULL); |
| 755 | /*cache_inval_vp(vp, CINV_DESTROY);*/ |
| 756 | error = 0; |
| 757 | |
| 758 | out: |
| 759 | vrele(vp); |
| 760 | |
| 761 | return error; |
| 762 | } |
| 763 | |
| 764 | /* --------------------------------------------------------------------- */ |
| 765 | |
| 766 | static int |
| 767 | tmpfs_nlink(struct vop_nlink_args *v) |
| 768 | { |
| 769 | struct vnode *dvp = v->a_dvp; |
| 770 | struct vnode *vp = v->a_vp; |
| 771 | struct namecache *ncp = v->a_nch->ncp; |
| 772 | struct tmpfs_dirent *de; |
| 773 | struct tmpfs_node *node; |
| 774 | struct tmpfs_node *dnode; |
| 775 | int error; |
| 776 | |
| 777 | KKASSERT(dvp != vp); /* XXX When can this be false? */ |
| 778 | |
| 779 | node = VP_TO_TMPFS_NODE(vp); |
| 780 | dnode = VP_TO_TMPFS_NODE(dvp); |
| 781 | |
| 782 | /* XXX: Why aren't the following two tests done by the caller? */ |
| 783 | |
| 784 | /* Hard links of directories are forbidden. */ |
| 785 | if (vp->v_type == VDIR) { |
| 786 | error = EPERM; |
| 787 | goto out; |
| 788 | } |
| 789 | |
| 790 | /* Cannot create cross-device links. */ |
| 791 | if (dvp->v_mount != vp->v_mount) { |
| 792 | error = EXDEV; |
| 793 | goto out; |
| 794 | } |
| 795 | |
| 796 | /* Ensure that we do not overflow the maximum number of links imposed |
| 797 | * by the system. */ |
| 798 | KKASSERT(node->tn_links <= LINK_MAX); |
| 799 | if (node->tn_links == LINK_MAX) { |
| 800 | error = EMLINK; |
| 801 | goto out; |
| 802 | } |
| 803 | |
| 804 | /* We cannot create links of files marked immutable or append-only. */ |
| 805 | if (node->tn_flags & (IMMUTABLE | APPEND)) { |
| 806 | error = EPERM; |
| 807 | goto out; |
| 808 | } |
| 809 | |
| 810 | /* Allocate a new directory entry to represent the node. */ |
| 811 | error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, |
| 812 | ncp->nc_name, ncp->nc_nlen, &de); |
| 813 | if (error != 0) |
| 814 | goto out; |
| 815 | |
| 816 | /* Insert the new directory entry into the appropriate directory. */ |
| 817 | tmpfs_dir_attach(dnode, de); |
| 818 | |
| 819 | /* vp link count has changed, so update node times. */ |
| 820 | |
| 821 | TMPFS_NODE_LOCK(node); |
| 822 | node->tn_status |= TMPFS_NODE_CHANGED; |
| 823 | TMPFS_NODE_UNLOCK(node); |
| 824 | tmpfs_update(vp); |
| 825 | |
| 826 | cache_setunresolved(v->a_nch); |
| 827 | cache_setvp(v->a_nch, vp); |
| 828 | error = 0; |
| 829 | |
| 830 | out: |
| 831 | return error; |
| 832 | } |
| 833 | |
| 834 | /* --------------------------------------------------------------------- */ |
| 835 | |
| 836 | static int |
| 837 | tmpfs_nrename(struct vop_nrename_args *v) |
| 838 | { |
| 839 | struct vnode *fdvp = v->a_fdvp; |
| 840 | struct namecache *fncp = v->a_fnch->ncp; |
| 841 | struct vnode *fvp = fncp->nc_vp; |
| 842 | struct vnode *tdvp = v->a_tdvp; |
| 843 | struct namecache *tncp = v->a_tnch->ncp; |
| 844 | struct vnode *tvp = tncp->nc_vp; |
| 845 | struct tmpfs_dirent *de; |
| 846 | struct tmpfs_mount *tmp; |
| 847 | struct tmpfs_node *fdnode; |
| 848 | struct tmpfs_node *fnode; |
| 849 | struct tmpfs_node *tnode; |
| 850 | struct tmpfs_node *tdnode; |
| 851 | char *newname; |
| 852 | char *oldname; |
| 853 | int error; |
| 854 | |
| 855 | tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp); |
| 856 | |
| 857 | /* Disallow cross-device renames. |
| 858 | * XXX Why isn't this done by the caller? */ |
| 859 | if (fvp->v_mount != tdvp->v_mount || |
| 860 | (tvp != NULL && fvp->v_mount != tvp->v_mount)) { |
| 861 | error = EXDEV; |
| 862 | goto out; |
| 863 | } |
| 864 | |
| 865 | tmp = VFS_TO_TMPFS(tdvp->v_mount); |
| 866 | tdnode = VP_TO_TMPFS_DIR(tdvp); |
| 867 | |
| 868 | /* If source and target are the same file, there is nothing to do. */ |
| 869 | if (fvp == tvp) { |
| 870 | error = 0; |
| 871 | goto out; |
| 872 | } |
| 873 | |
| 874 | fdnode = VP_TO_TMPFS_DIR(fdvp); |
| 875 | fnode = VP_TO_TMPFS_NODE(fvp); |
| 876 | de = tmpfs_dir_lookup(fdnode, fnode, fncp); |
| 877 | |
| 878 | /* Avoid manipulating '.' and '..' entries. */ |
| 879 | if (de == NULL) { |
| 880 | error = ENOENT; |
| 881 | goto out_locked; |
| 882 | } |
| 883 | KKASSERT(de->td_node == fnode); |
| 884 | |
| 885 | /* |
| 886 | * If replacing an entry in the target directory and that entry |
| 887 | * is a directory, it must be empty. |
| 888 | * |
| 889 | * Kern_rename gurantees the destination to be a directory |
| 890 | * if the source is one (it does?). |
| 891 | */ |
| 892 | if (tvp != NULL) { |
| 893 | KKASSERT(tnode != NULL); |
| 894 | |
| 895 | if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || |
| 896 | (tdnode->tn_flags & (APPEND | IMMUTABLE))) { |
| 897 | error = EPERM; |
| 898 | goto out_locked; |
| 899 | } |
| 900 | |
| 901 | if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { |
| 902 | if (tnode->tn_size > 0) { |
| 903 | error = ENOTEMPTY; |
| 904 | goto out_locked; |
| 905 | } |
| 906 | } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { |
| 907 | error = ENOTDIR; |
| 908 | goto out_locked; |
| 909 | } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { |
| 910 | error = EISDIR; |
| 911 | goto out_locked; |
| 912 | } else { |
| 913 | KKASSERT(fnode->tn_type != VDIR && |
| 914 | tnode->tn_type != VDIR); |
| 915 | } |
| 916 | } |
| 917 | |
| 918 | if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || |
| 919 | (fdnode->tn_flags & (APPEND | IMMUTABLE))) { |
| 920 | error = EPERM; |
| 921 | goto out_locked; |
| 922 | } |
| 923 | |
| 924 | /* |
| 925 | * Ensure that we have enough memory to hold the new name, if it |
| 926 | * has to be changed. |
| 927 | */ |
| 928 | if (fncp->nc_nlen != tncp->nc_nlen || |
| 929 | bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) { |
| 930 | newname = kmalloc(tncp->nc_nlen + 1, M_TMPFSNAME, M_WAITOK); |
| 931 | bcopy(tncp->nc_name, newname, tncp->nc_nlen); |
| 932 | newname[tncp->nc_nlen] = '\0'; |
| 933 | } else { |
| 934 | newname = NULL; |
| 935 | } |
| 936 | |
| 937 | /* |
| 938 | * Unlink entry from source directory. Note that the kernel has |
| 939 | * already checked for illegal recursion cases (renaming a directory |
| 940 | * into a subdirectory of itself). |
| 941 | */ |
| 942 | if (fdnode != tdnode) |
| 943 | tmpfs_dir_detach(fdnode, de); |
| 944 | |
| 945 | /* |
| 946 | * Handle any name change. Swap with newname, we will |
| 947 | * deallocate it at the end. |
| 948 | */ |
| 949 | if (newname != NULL) { |
| 950 | #if 0 |
| 951 | TMPFS_NODE_LOCK(fnode); |
| 952 | fnode->tn_status |= TMPFS_NODE_CHANGED; |
| 953 | TMPFS_NODE_UNLOCK(fnode); |
| 954 | #endif |
| 955 | oldname = de->td_name; |
| 956 | de->td_name = newname; |
| 957 | de->td_namelen = (uint16_t)tncp->nc_nlen; |
| 958 | newname = oldname; |
| 959 | } |
| 960 | |
| 961 | /* |
| 962 | * Link entry to target directory. If the entry |
| 963 | * represents a directory move the parent linkage |
| 964 | * as well. |
| 965 | */ |
| 966 | if (fdnode != tdnode) { |
| 967 | if (de->td_node->tn_type == VDIR) { |
| 968 | TMPFS_VALIDATE_DIR(fnode); |
| 969 | |
| 970 | TMPFS_NODE_LOCK(tdnode); |
| 971 | tdnode->tn_links++; |
| 972 | tdnode->tn_status |= TMPFS_NODE_MODIFIED; |
| 973 | TMPFS_NODE_UNLOCK(tdnode); |
| 974 | |
| 975 | TMPFS_NODE_LOCK(fnode); |
| 976 | fnode->tn_dir.tn_parent = tdnode; |
| 977 | fnode->tn_status |= TMPFS_NODE_CHANGED; |
| 978 | TMPFS_NODE_UNLOCK(fnode); |
| 979 | |
| 980 | TMPFS_NODE_LOCK(fdnode); |
| 981 | fdnode->tn_links--; |
| 982 | fdnode->tn_status |= TMPFS_NODE_MODIFIED; |
| 983 | TMPFS_NODE_UNLOCK(fdnode); |
| 984 | } |
| 985 | tmpfs_dir_attach(tdnode, de); |
| 986 | } else { |
| 987 | TMPFS_NODE_LOCK(tdnode); |
| 988 | tdnode->tn_status |= TMPFS_NODE_MODIFIED; |
| 989 | TMPFS_NODE_UNLOCK(tdnode); |
| 990 | } |
| 991 | |
| 992 | /* |
| 993 | * If we are overwriting an entry, we have to remove the old one |
| 994 | * from the target directory. |
| 995 | */ |
| 996 | if (tvp != NULL) { |
| 997 | /* Remove the old entry from the target directory. */ |
| 998 | de = tmpfs_dir_lookup(tdnode, tnode, tncp); |
| 999 | tmpfs_dir_detach(tdnode, de); |
| 1000 | |
| 1001 | /* |
| 1002 | * Free the directory entry we just deleted. Note that the |
| 1003 | * node referred by it will not be removed until the vnode is |
| 1004 | * really reclaimed. |
| 1005 | */ |
| 1006 | tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de); |
| 1007 | /*cache_inval_vp(tvp, CINV_DESTROY);*/ |
| 1008 | } |
| 1009 | |
| 1010 | /* |
| 1011 | * Finish up |
| 1012 | */ |
| 1013 | if (newname) { |
| 1014 | kfree(newname, M_TMPFSNAME); |
| 1015 | newname = NULL; |
| 1016 | } |
| 1017 | cache_rename(v->a_fnch, v->a_tnch); |
| 1018 | error = 0; |
| 1019 | |
| 1020 | out_locked: |
| 1021 | ; |
| 1022 | |
| 1023 | out: |
| 1024 | /* Release target nodes. */ |
| 1025 | /* XXX: I don't understand when tdvp can be the same as tvp, but |
| 1026 | * other code takes care of this... */ |
| 1027 | if (tdvp == tvp) |
| 1028 | vrele(tdvp); |
| 1029 | |
| 1030 | return error; |
| 1031 | } |
| 1032 | |
| 1033 | /* --------------------------------------------------------------------- */ |
| 1034 | |
| 1035 | static int |
| 1036 | tmpfs_nmkdir(struct vop_nmkdir_args *v) |
| 1037 | { |
| 1038 | struct vnode *dvp = v->a_dvp; |
| 1039 | struct vnode **vpp = v->a_vpp; |
| 1040 | struct namecache *ncp = v->a_nch->ncp; |
| 1041 | struct vattr *vap = v->a_vap; |
| 1042 | struct ucred *cred = v->a_cred; |
| 1043 | int error; |
| 1044 | |
| 1045 | KKASSERT(vap->va_type == VDIR); |
| 1046 | |
| 1047 | error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); |
| 1048 | if (error == 0) { |
| 1049 | cache_setunresolved(v->a_nch); |
| 1050 | cache_setvp(v->a_nch, *vpp); |
| 1051 | } |
| 1052 | |
| 1053 | return error; |
| 1054 | } |
| 1055 | |
| 1056 | /* --------------------------------------------------------------------- */ |
| 1057 | |
| 1058 | static int |
| 1059 | tmpfs_nrmdir(struct vop_nrmdir_args *v) |
| 1060 | { |
| 1061 | struct vnode *dvp = v->a_dvp; |
| 1062 | struct namecache *ncp = v->a_nch->ncp; |
| 1063 | struct vnode *vp; |
| 1064 | struct tmpfs_dirent *de; |
| 1065 | struct tmpfs_mount *tmp; |
| 1066 | struct tmpfs_node *dnode; |
| 1067 | struct tmpfs_node *node; |
| 1068 | int error; |
| 1069 | |
| 1070 | /* |
| 1071 | * We have to acquire the vp from v->a_nch because |
| 1072 | * we will likely unresolve the namecache entry, and |
| 1073 | * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim |
| 1074 | * sequence. |
| 1075 | */ |
| 1076 | error = cache_vref(v->a_nch, v->a_cred, &vp); |
| 1077 | KKASSERT(error == 0); |
| 1078 | |
| 1079 | /* |
| 1080 | * Prevalidate so we don't hit an assertion later |
| 1081 | */ |
| 1082 | if (vp->v_type != VDIR) { |
| 1083 | error = ENOTDIR; |
| 1084 | goto out; |
| 1085 | } |
| 1086 | |
| 1087 | tmp = VFS_TO_TMPFS(dvp->v_mount); |
| 1088 | dnode = VP_TO_TMPFS_DIR(dvp); |
| 1089 | node = VP_TO_TMPFS_DIR(vp); |
| 1090 | |
| 1091 | /* Directories with more than two entries ('.' and '..') cannot be |
| 1092 | * removed. */ |
| 1093 | if (node->tn_size > 0) { |
| 1094 | error = ENOTEMPTY; |
| 1095 | goto out; |
| 1096 | } |
| 1097 | |
| 1098 | if ((dnode->tn_flags & APPEND) |
| 1099 | || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { |
| 1100 | error = EPERM; |
| 1101 | goto out; |
| 1102 | } |
| 1103 | |
| 1104 | /* This invariant holds only if we are not trying to remove "..". |
| 1105 | * We checked for that above so this is safe now. */ |
| 1106 | KKASSERT(node->tn_dir.tn_parent == dnode); |
| 1107 | |
| 1108 | /* Get the directory entry associated with node (vp). This was |
| 1109 | * filled by tmpfs_lookup while looking up the entry. */ |
| 1110 | de = tmpfs_dir_lookup(dnode, node, ncp); |
| 1111 | KKASSERT(TMPFS_DIRENT_MATCHES(de, |
| 1112 | ncp->nc_name, |
| 1113 | ncp->nc_nlen)); |
| 1114 | |
| 1115 | /* Check flags to see if we are allowed to remove the directory. */ |
| 1116 | if ((dnode->tn_flags & APPEND) || |
| 1117 | node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { |
| 1118 | error = EPERM; |
| 1119 | goto out; |
| 1120 | } |
| 1121 | |
| 1122 | |
| 1123 | /* Detach the directory entry from the directory (dnode). */ |
| 1124 | tmpfs_dir_detach(dnode, de); |
| 1125 | |
| 1126 | /* No vnode should be allocated for this entry from this point */ |
| 1127 | TMPFS_NODE_LOCK(node); |
| 1128 | TMPFS_ASSERT_ELOCKED(node); |
| 1129 | TMPFS_NODE_LOCK(dnode); |
| 1130 | TMPFS_ASSERT_ELOCKED(dnode); |
| 1131 | |
| 1132 | #if 0 |
| 1133 | /* handled by tmpfs_free_node */ |
| 1134 | KKASSERT(node->tn_links > 0); |
| 1135 | node->tn_links--; |
| 1136 | node->tn_dir.tn_parent = NULL; |
| 1137 | #endif |
| 1138 | node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ |
| 1139 | TMPFS_NODE_MODIFIED; |
| 1140 | |
| 1141 | #if 0 |
| 1142 | /* handled by tmpfs_free_node */ |
| 1143 | KKASSERT(dnode->tn_links > 0); |
| 1144 | dnode->tn_links--; |
| 1145 | #endif |
| 1146 | dnode->tn_status |= TMPFS_NODE_ACCESSED | \ |
| 1147 | TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; |
| 1148 | |
| 1149 | TMPFS_NODE_UNLOCK(dnode); |
| 1150 | TMPFS_NODE_UNLOCK(node); |
| 1151 | |
| 1152 | /* Free the directory entry we just deleted. Note that the node |
| 1153 | * referred by it will not be removed until the vnode is really |
| 1154 | * reclaimed. */ |
| 1155 | tmpfs_free_dirent(tmp, de); |
| 1156 | |
| 1157 | /* Release the deleted vnode (will destroy the node, notify |
| 1158 | * interested parties and clean it from the cache). */ |
| 1159 | |
| 1160 | TMPFS_NODE_LOCK(dnode); |
| 1161 | dnode->tn_status |= TMPFS_NODE_CHANGED; |
| 1162 | TMPFS_NODE_UNLOCK(dnode); |
| 1163 | tmpfs_update(dvp); |
| 1164 | |
| 1165 | cache_setunresolved(v->a_nch); |
| 1166 | cache_setvp(v->a_nch, NULL); |
| 1167 | /*cache_inval_vp(vp, CINV_DESTROY);*/ |
| 1168 | error = 0; |
| 1169 | |
| 1170 | out: |
| 1171 | vrele(vp); |
| 1172 | |
| 1173 | return error; |
| 1174 | } |
| 1175 | |
| 1176 | /* --------------------------------------------------------------------- */ |
| 1177 | |
| 1178 | static int |
| 1179 | tmpfs_nsymlink(struct vop_nsymlink_args *v) |
| 1180 | { |
| 1181 | struct vnode *dvp = v->a_dvp; |
| 1182 | struct vnode **vpp = v->a_vpp; |
| 1183 | struct namecache *ncp = v->a_nch->ncp; |
| 1184 | struct vattr *vap = v->a_vap; |
| 1185 | struct ucred *cred = v->a_cred; |
| 1186 | char *target = v->a_target; |
| 1187 | int error; |
| 1188 | |
| 1189 | vap->va_type = VLNK; |
| 1190 | error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target); |
| 1191 | if (error == 0) { |
| 1192 | cache_setunresolved(v->a_nch); |
| 1193 | cache_setvp(v->a_nch, *vpp); |
| 1194 | } |
| 1195 | |
| 1196 | return error; |
| 1197 | } |
| 1198 | |
| 1199 | /* --------------------------------------------------------------------- */ |
| 1200 | |
| 1201 | static int |
| 1202 | tmpfs_readdir(struct vop_readdir_args *v) |
| 1203 | { |
| 1204 | struct vnode *vp = v->a_vp; |
| 1205 | struct uio *uio = v->a_uio; |
| 1206 | int *eofflag = v->a_eofflag; |
| 1207 | off_t **cookies = v->a_cookies; |
| 1208 | int *ncookies = v->a_ncookies; |
| 1209 | struct tmpfs_mount *tmp; |
| 1210 | int error; |
| 1211 | off_t startoff; |
| 1212 | off_t cnt = 0; |
| 1213 | struct tmpfs_node *node; |
| 1214 | |
| 1215 | /* This operation only makes sense on directory nodes. */ |
| 1216 | if (vp->v_type != VDIR) |
| 1217 | return ENOTDIR; |
| 1218 | |
| 1219 | tmp = VFS_TO_TMPFS(vp->v_mount); |
| 1220 | node = VP_TO_TMPFS_DIR(vp); |
| 1221 | startoff = uio->uio_offset; |
| 1222 | |
| 1223 | if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) { |
| 1224 | error = tmpfs_dir_getdotdent(node, uio); |
| 1225 | if (error != 0) |
| 1226 | goto outok; |
| 1227 | cnt++; |
| 1228 | } |
| 1229 | |
| 1230 | if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) { |
| 1231 | error = tmpfs_dir_getdotdotdent(tmp, node, uio); |
| 1232 | if (error != 0) |
| 1233 | goto outok; |
| 1234 | cnt++; |
| 1235 | } |
| 1236 | |
| 1237 | error = tmpfs_dir_getdents(node, uio, &cnt); |
| 1238 | |
| 1239 | outok: |
| 1240 | KKASSERT(error >= -1); |
| 1241 | |
| 1242 | if (error == -1) |
| 1243 | error = 0; |
| 1244 | |
| 1245 | if (eofflag != NULL) |
| 1246 | *eofflag = |
| 1247 | (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); |
| 1248 | |
| 1249 | /* Update NFS-related variables. */ |
| 1250 | if (error == 0 && cookies != NULL && ncookies != NULL) { |
| 1251 | off_t i; |
| 1252 | off_t off = startoff; |
| 1253 | struct tmpfs_dirent *de = NULL; |
| 1254 | |
| 1255 | *ncookies = cnt; |
| 1256 | *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK); |
| 1257 | |
| 1258 | for (i = 0; i < cnt; i++) { |
| 1259 | KKASSERT(off != TMPFS_DIRCOOKIE_EOF); |
| 1260 | if (off == TMPFS_DIRCOOKIE_DOT) { |
| 1261 | off = TMPFS_DIRCOOKIE_DOTDOT; |
| 1262 | } else { |
| 1263 | if (off == TMPFS_DIRCOOKIE_DOTDOT) { |
| 1264 | de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); |
| 1265 | } else if (de != NULL) { |
| 1266 | de = TAILQ_NEXT(de, td_entries); |
| 1267 | } else { |
| 1268 | de = tmpfs_dir_lookupbycookie(node, |
| 1269 | off); |
| 1270 | KKASSERT(de != NULL); |
| 1271 | de = TAILQ_NEXT(de, td_entries); |
| 1272 | } |
| 1273 | if (de == NULL) |
| 1274 | off = TMPFS_DIRCOOKIE_EOF; |
| 1275 | else |
| 1276 | off = tmpfs_dircookie(de); |
| 1277 | } |
| 1278 | |
| 1279 | (*cookies)[i] = off; |
| 1280 | } |
| 1281 | KKASSERT(uio->uio_offset == off); |
| 1282 | } |
| 1283 | |
| 1284 | return error; |
| 1285 | } |
| 1286 | |
| 1287 | /* --------------------------------------------------------------------- */ |
| 1288 | |
| 1289 | static int |
| 1290 | tmpfs_readlink(struct vop_readlink_args *v) |
| 1291 | { |
| 1292 | struct vnode *vp = v->a_vp; |
| 1293 | struct uio *uio = v->a_uio; |
| 1294 | |
| 1295 | int error; |
| 1296 | struct tmpfs_node *node; |
| 1297 | |
| 1298 | KKASSERT(uio->uio_offset == 0); |
| 1299 | KKASSERT(vp->v_type == VLNK); |
| 1300 | |
| 1301 | node = VP_TO_TMPFS_NODE(vp); |
| 1302 | |
| 1303 | error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid), |
| 1304 | uio); |
| 1305 | TMPFS_NODE_LOCK(node); |
| 1306 | node->tn_status |= TMPFS_NODE_ACCESSED; |
| 1307 | TMPFS_NODE_UNLOCK(node); |
| 1308 | |
| 1309 | return error; |
| 1310 | } |
| 1311 | |
| 1312 | /* --------------------------------------------------------------------- */ |
| 1313 | |
| 1314 | static int |
| 1315 | tmpfs_inactive(struct vop_inactive_args *v) |
| 1316 | { |
| 1317 | struct vnode *vp = v->a_vp; |
| 1318 | |
| 1319 | struct tmpfs_node *node; |
| 1320 | |
| 1321 | node = VP_TO_TMPFS_NODE(vp); |
| 1322 | |
| 1323 | /* |
| 1324 | * Get rid of unreferenced deleted vnodes sooner rather than |
| 1325 | * later so the data memory can be recovered immediately. |
| 1326 | * |
| 1327 | * We must truncate the vnode to prevent the normal reclamation |
| 1328 | * path from flushing the data for the removed file to disk. |
| 1329 | */ |
| 1330 | TMPFS_NODE_LOCK(node); |
| 1331 | if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && |
| 1332 | (node->tn_links == 0 || |
| 1333 | (node->tn_links == 1 && node->tn_type == VDIR && |
| 1334 | node->tn_dir.tn_parent))) |
| 1335 | { |
| 1336 | node->tn_vpstate = TMPFS_VNODE_DOOMED; |
| 1337 | TMPFS_NODE_UNLOCK(node); |
| 1338 | if (node->tn_type == VREG) |
| 1339 | tmpfs_truncate(vp, 0); |
| 1340 | vrecycle(vp); |
| 1341 | } else { |
| 1342 | TMPFS_NODE_UNLOCK(node); |
| 1343 | } |
| 1344 | |
| 1345 | return 0; |
| 1346 | } |
| 1347 | |
| 1348 | /* --------------------------------------------------------------------- */ |
| 1349 | |
| 1350 | int |
| 1351 | tmpfs_reclaim(struct vop_reclaim_args *v) |
| 1352 | { |
| 1353 | struct vnode *vp = v->a_vp; |
| 1354 | struct tmpfs_mount *tmp; |
| 1355 | struct tmpfs_node *node; |
| 1356 | |
| 1357 | node = VP_TO_TMPFS_NODE(vp); |
| 1358 | tmp = VFS_TO_TMPFS(vp->v_mount); |
| 1359 | |
| 1360 | tmpfs_free_vp(vp); |
| 1361 | |
| 1362 | /* |
| 1363 | * If the node referenced by this vnode was deleted by the |
| 1364 | * user, we must free its associated data structures now that |
| 1365 | * the vnode is being reclaimed. |
| 1366 | * |
| 1367 | * Directories have an extra link ref. |
| 1368 | */ |
| 1369 | TMPFS_NODE_LOCK(node); |
| 1370 | if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && |
| 1371 | (node->tn_links == 0 || |
| 1372 | (node->tn_links == 1 && node->tn_type == VDIR && |
| 1373 | node->tn_dir.tn_parent))) |
| 1374 | { |
| 1375 | node->tn_vpstate = TMPFS_VNODE_DOOMED; |
| 1376 | tmpfs_free_node(tmp, node); |
| 1377 | /* eats the lock */ |
| 1378 | } else { |
| 1379 | TMPFS_NODE_UNLOCK(node); |
| 1380 | } |
| 1381 | |
| 1382 | KKASSERT(vp->v_data == NULL); |
| 1383 | return 0; |
| 1384 | } |
| 1385 | |
| 1386 | /* --------------------------------------------------------------------- */ |
| 1387 | |
| 1388 | static int |
| 1389 | tmpfs_print(struct vop_print_args *v) |
| 1390 | { |
| 1391 | struct vnode *vp = v->a_vp; |
| 1392 | |
| 1393 | struct tmpfs_node *node; |
| 1394 | |
| 1395 | node = VP_TO_TMPFS_NODE(vp); |
| 1396 | |
| 1397 | kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n", |
| 1398 | node, node->tn_flags, node->tn_links); |
| 1399 | kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n", |
| 1400 | node->tn_mode, node->tn_uid, node->tn_gid, |
| 1401 | (uintmax_t)node->tn_size, node->tn_status); |
| 1402 | |
| 1403 | if (vp->v_type == VFIFO) |
| 1404 | fifo_printinfo(vp); |
| 1405 | |
| 1406 | kprintf("\n"); |
| 1407 | |
| 1408 | return 0; |
| 1409 | } |
| 1410 | |
| 1411 | /* --------------------------------------------------------------------- */ |
| 1412 | |
| 1413 | static int |
| 1414 | tmpfs_pathconf(struct vop_pathconf_args *v) |
| 1415 | { |
| 1416 | int name = v->a_name; |
| 1417 | register_t *retval = v->a_retval; |
| 1418 | |
| 1419 | int error; |
| 1420 | |
| 1421 | error = 0; |
| 1422 | |
| 1423 | switch (name) { |
| 1424 | case _PC_LINK_MAX: |
| 1425 | *retval = LINK_MAX; |
| 1426 | break; |
| 1427 | |
| 1428 | case _PC_NAME_MAX: |
| 1429 | *retval = NAME_MAX; |
| 1430 | break; |
| 1431 | |
| 1432 | case _PC_PATH_MAX: |
| 1433 | *retval = PATH_MAX; |
| 1434 | break; |
| 1435 | |
| 1436 | case _PC_PIPE_BUF: |
| 1437 | *retval = PIPE_BUF; |
| 1438 | break; |
| 1439 | |
| 1440 | case _PC_CHOWN_RESTRICTED: |
| 1441 | *retval = 1; |
| 1442 | break; |
| 1443 | |
| 1444 | case _PC_NO_TRUNC: |
| 1445 | *retval = 1; |
| 1446 | break; |
| 1447 | |
| 1448 | case _PC_SYNC_IO: |
| 1449 | *retval = 1; |
| 1450 | break; |
| 1451 | |
| 1452 | case _PC_FILESIZEBITS: |
| 1453 | *retval = 0; /* XXX Don't know which value should I return. */ |
| 1454 | break; |
| 1455 | |
| 1456 | default: |
| 1457 | error = EINVAL; |
| 1458 | } |
| 1459 | |
| 1460 | return error; |
| 1461 | } |
| 1462 | |
| 1463 | /* --------------------------------------------------------------------- */ |
| 1464 | |
| 1465 | /* |
| 1466 | * vnode operations vector used for files stored in a tmpfs file system. |
| 1467 | */ |
| 1468 | struct vop_ops tmpfs_vnode_vops = { |
| 1469 | .vop_default = vop_defaultop, |
| 1470 | .vop_getpages = vop_stdgetpages, |
| 1471 | .vop_putpages = vop_stdputpages, |
| 1472 | .vop_ncreate = tmpfs_ncreate, |
| 1473 | .vop_nresolve = tmpfs_nresolve, |
| 1474 | .vop_nlookupdotdot = tmpfs_nlookupdotdot, |
| 1475 | .vop_nmknod = tmpfs_nmknod, |
| 1476 | .vop_open = tmpfs_open, |
| 1477 | .vop_close = tmpfs_close, |
| 1478 | .vop_access = tmpfs_access, |
| 1479 | .vop_getattr = tmpfs_getattr, |
| 1480 | .vop_setattr = tmpfs_setattr, |
| 1481 | .vop_read = tmpfs_read, |
| 1482 | .vop_write = tmpfs_write, |
| 1483 | .vop_fsync = tmpfs_fsync, |
| 1484 | .vop_nremove = tmpfs_nremove, |
| 1485 | .vop_nlink = tmpfs_nlink, |
| 1486 | .vop_nrename = tmpfs_nrename, |
| 1487 | .vop_nmkdir = tmpfs_nmkdir, |
| 1488 | .vop_nrmdir = tmpfs_nrmdir, |
| 1489 | .vop_nsymlink = tmpfs_nsymlink, |
| 1490 | .vop_readdir = tmpfs_readdir, |
| 1491 | .vop_readlink = tmpfs_readlink, |
| 1492 | .vop_inactive = tmpfs_inactive, |
| 1493 | .vop_reclaim = tmpfs_reclaim, |
| 1494 | .vop_print = tmpfs_print, |
| 1495 | .vop_pathconf = tmpfs_pathconf, |
| 1496 | .vop_bmap = tmpfs_bmap, |
| 1497 | .vop_strategy = tmpfs_strategy, |
| 1498 | .vop_advlock = tmpfs_advlock, |
| 1499 | }; |