1 /* 2 * Copyright (c) 2007 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.3 2007/11/19 00:53:40 dillon Exp $ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/fcntl.h> 41 #include <sys/namecache.h> 42 #include <sys/vnode.h> 43 #include <sys/lockf.h> 44 #include <sys/event.h> 45 #include <sys/stat.h> 46 #include "hammer.h" 47 48 /* 49 * USERFS VNOPS 50 */ 51 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/ 52 static int hammer_vop_fsync(struct vop_fsync_args *); 53 static int hammer_vop_read(struct vop_read_args *); 54 static int hammer_vop_write(struct vop_write_args *); 55 static int hammer_vop_access(struct vop_access_args *); 56 static int hammer_vop_advlock(struct vop_advlock_args *); 57 static int hammer_vop_close(struct vop_close_args *); 58 static int hammer_vop_ncreate(struct vop_ncreate_args *); 59 static int hammer_vop_getattr(struct vop_getattr_args *); 60 static int hammer_vop_nresolve(struct vop_nresolve_args *); 61 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *); 62 static int hammer_vop_nlink(struct vop_nlink_args *); 63 static int hammer_vop_nmkdir(struct vop_nmkdir_args *); 64 static int hammer_vop_nmknod(struct vop_nmknod_args *); 65 static int hammer_vop_open(struct vop_open_args *); 66 static int hammer_vop_pathconf(struct vop_pathconf_args *); 67 static int hammer_vop_print(struct vop_print_args *); 68 static int hammer_vop_readdir(struct vop_readdir_args *); 69 static int hammer_vop_readlink(struct vop_readlink_args *); 70 static int hammer_vop_nremove(struct vop_nremove_args *); 71 static int hammer_vop_nrename(struct vop_nrename_args *); 72 static int hammer_vop_nrmdir(struct vop_nrmdir_args *); 73 static int hammer_vop_setattr(struct vop_setattr_args *); 74 static int hammer_vop_strategy(struct vop_strategy_args *); 75 static int hammer_vop_nsymlink(struct vop_nsymlink_args *); 76 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *); 77 78 struct vop_ops hammer_vnode_vops = { 79 .vop_default = vop_defaultop, 80 .vop_fsync = hammer_vop_fsync, 81 .vop_read = hammer_vop_read, 82 .vop_write = hammer_vop_write, 83 .vop_access = hammer_vop_access, 84 .vop_advlock = hammer_vop_advlock, 85 .vop_close = hammer_vop_close, 86 .vop_ncreate = hammer_vop_ncreate, 87 .vop_getattr = hammer_vop_getattr, 88 .vop_inactive = hammer_vop_inactive, 89 .vop_reclaim = hammer_vop_reclaim, 90 .vop_nresolve = hammer_vop_nresolve, 91 .vop_nlookupdotdot = hammer_vop_nlookupdotdot, 92 .vop_nlink = hammer_vop_nlink, 93 .vop_nmkdir = hammer_vop_nmkdir, 94 .vop_nmknod = hammer_vop_nmknod, 95 .vop_open = hammer_vop_open, 96 .vop_pathconf = hammer_vop_pathconf, 97 .vop_print = hammer_vop_print, 98 .vop_readdir = hammer_vop_readdir, 99 .vop_readlink = hammer_vop_readlink, 100 .vop_nremove = hammer_vop_nremove, 101 .vop_nrename = hammer_vop_nrename, 102 .vop_nrmdir = hammer_vop_nrmdir, 103 .vop_setattr = hammer_vop_setattr, 104 .vop_strategy = hammer_vop_strategy, 105 .vop_nsymlink = hammer_vop_nsymlink, 106 .vop_nwhiteout = hammer_vop_nwhiteout 107 }; 108 109 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp, 110 struct ucred *cred, int flags); 111 static int hammer_vop_strategy_read(struct vop_strategy_args *ap); 112 static int hammer_vop_strategy_write(struct vop_strategy_args *ap); 113 114 #if 0 115 static 116 int 117 hammer_vop_vnoperate(struct vop_generic_args *) 118 { 119 return (VOCALL(&hammer_vnode_vops, ap)); 120 } 121 #endif 122 123 /* 124 * hammer_vop_fsync { vp, waitfor } 125 */ 126 static 127 int 128 hammer_vop_fsync(struct vop_fsync_args *ap) 129 { 130 return EOPNOTSUPP; 131 } 132 133 /* 134 * hammer_vop_read { vp, uio, ioflag, cred } 135 */ 136 static 137 int 138 hammer_vop_read(struct vop_read_args *ap) 139 { 140 struct hammer_transaction trans; 141 struct hammer_inode *ip; 142 off_t offset; 143 struct buf *bp; 144 struct uio *uio; 145 int error; 146 int n; 147 int seqcount; 148 149 if (ap->a_vp->v_type != VREG) 150 return (EINVAL); 151 ip = VTOI(ap->a_vp); 152 error = 0; 153 seqcount = ap->a_ioflag >> 16; 154 155 hammer_start_transaction(&trans, ip->hmp); 156 157 /* 158 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache. 159 */ 160 uio = ap->a_uio; 161 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) { 162 offset = uio->uio_offset & HAMMER_BUFMASK; 163 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size, 164 uio->uio_offset - offset, HAMMER_BUFSIZE, 165 MAXBSIZE, seqcount, &bp); 166 if (error) { 167 brelse(bp); 168 break; 169 } 170 bp->b_flags |= B_CLUSTEROK; 171 n = HAMMER_BUFSIZE - offset; 172 if (n > uio->uio_resid) 173 n = uio->uio_resid; 174 if (n > ip->ino_rec.ino_size - uio->uio_offset) 175 n = (int)(ip->ino_rec.ino_size - uio->uio_offset); 176 error = uiomove((char *)bp->b_data + offset, n, uio); 177 if (error) { 178 bqrelse(bp); 179 break; 180 } 181 ip->ino_rec.ino_atime = trans.tid; 182 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES); 183 bqrelse(bp); 184 } 185 hammer_commit_transaction(&trans); 186 return (error); 187 } 188 189 /* 190 * hammer_vop_write { vp, uio, ioflag, cred } 191 */ 192 static 193 int 194 hammer_vop_write(struct vop_write_args *ap) 195 { 196 struct hammer_transaction trans; 197 struct hammer_inode *ip; 198 struct uio *uio; 199 off_t offset; 200 struct buf *bp; 201 int error; 202 int n; 203 204 if (ap->a_vp->v_type != VREG) 205 return (EINVAL); 206 ip = VTOI(ap->a_vp); 207 error = 0; 208 209 /* 210 * Create a transaction to cover the operations we perform. 211 */ 212 hammer_start_transaction(&trans, ip->hmp); 213 uio = ap->a_uio; 214 215 /* 216 * Check append mode 217 */ 218 if (ap->a_ioflag & IO_APPEND) 219 uio->uio_offset = ip->ino_rec.ino_size; 220 221 /* 222 * Check for illegal write offsets. Valid range is 0...2^63-1 223 */ 224 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) 225 return (EFBIG); 226 227 /* 228 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache. 229 */ 230 while (uio->uio_resid > 0) { 231 offset = uio->uio_offset & HAMMER_BUFMASK; 232 if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) { 233 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE, 234 0, 0); 235 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) { 236 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE, 237 0, 0); 238 vfs_bio_clrbuf(bp); 239 } else { 240 error = bread(ap->a_vp, uio->uio_offset - offset, 241 HAMMER_BUFSIZE, &bp); 242 if (error) { 243 brelse(bp); 244 break; 245 } 246 } 247 n = HAMMER_BUFSIZE - offset; 248 if (n > uio->uio_resid) 249 n = uio->uio_resid; 250 error = uiomove((char *)bp->b_data + offset, n, uio); 251 if (error) { 252 brelse(bp); 253 break; 254 } 255 bp->b_flags |= B_CLUSTEROK; 256 if (ip->ino_rec.ino_size < uio->uio_offset) { 257 ip->ino_rec.ino_size = uio->uio_offset; 258 ip->ino_rec.ino_mtime = trans.tid; 259 hammer_modify_inode(&trans, ip, 260 HAMMER_INODE_RDIRTY | HAMMER_INODE_ITIMES); 261 } 262 if (ap->a_ioflag & IO_SYNC) { 263 bwrite(bp); 264 } else if (ap->a_ioflag & IO_DIRECT) { 265 bawrite(bp); 266 } else { 267 bdwrite(bp); 268 } 269 } 270 if (error) 271 hammer_abort_transaction(&trans); 272 else 273 hammer_commit_transaction(&trans); 274 return (error); 275 } 276 277 /* 278 * hammer_vop_access { vp, mode, cred } 279 */ 280 static 281 int 282 hammer_vop_access(struct vop_access_args *ap) 283 { 284 struct hammer_inode *ip = VTOI(ap->a_vp); 285 uid_t uid; 286 gid_t gid; 287 int error; 288 289 uid = hammer_to_unix_xid(&ip->ino_data.uid); 290 gid = hammer_to_unix_xid(&ip->ino_data.gid); 291 292 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode, 293 ip->ino_data.uflags); 294 return (error); 295 } 296 297 /* 298 * hammer_vop_advlock { vp, id, op, fl, flags } 299 */ 300 static 301 int 302 hammer_vop_advlock(struct vop_advlock_args *ap) 303 { 304 struct hammer_inode *ip = VTOI(ap->a_vp); 305 306 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size)); 307 } 308 309 /* 310 * hammer_vop_close { vp, fflag } 311 */ 312 static 313 int 314 hammer_vop_close(struct vop_close_args *ap) 315 { 316 return EOPNOTSUPP; 317 } 318 319 /* 320 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap } 321 * 322 * The operating system has already ensured that the directory entry 323 * does not exist and done all appropriate namespace locking. 324 */ 325 static 326 int 327 hammer_vop_ncreate(struct vop_ncreate_args *ap) 328 { 329 struct hammer_transaction trans; 330 struct hammer_inode *dip; 331 struct hammer_inode *nip; 332 struct nchandle *nch; 333 int error; 334 335 nch = ap->a_nch; 336 dip = VTOI(ap->a_dvp); 337 338 /* 339 * Create a transaction to cover the operations we perform. 340 */ 341 hammer_start_transaction(&trans, dip->hmp); 342 343 /* 344 * Create a new filesystem object of the requested type. The 345 * returned inode will be referenceds but not locked. 346 */ 347 348 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip); 349 if (error) { 350 hammer_abort_transaction(&trans); 351 *ap->a_vpp = NULL; 352 return (error); 353 } 354 355 /* 356 * Add the new filesystem object to the directory. This will also 357 * bump the inode's link count. 358 */ 359 error = hammer_add_directory(&trans, dip, nch->ncp, nip); 360 361 /* 362 * Finish up. 363 */ 364 if (error) { 365 hammer_rel_inode(nip); 366 hammer_abort_transaction(&trans); 367 *ap->a_vpp = NULL; 368 } else { 369 hammer_commit_transaction(&trans); 370 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp); 371 hammer_rel_inode(nip); 372 } 373 return (error); 374 } 375 376 /* 377 * hammer_vop_getattr { vp, vap } 378 */ 379 static 380 int 381 hammer_vop_getattr(struct vop_getattr_args *ap) 382 { 383 struct hammer_inode *ip = VTOI(ap->a_vp); 384 struct vattr *vap = ap->a_vap; 385 386 #if 0 387 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) && 388 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 && 389 ip->obj_asof == 0 390 ) { 391 /* LAZYMOD XXX */ 392 } 393 hammer_itimes(ap->a_vp); 394 #endif 395 396 vap->va_fsid = ip->hmp->fsid_udev; 397 vap->va_fileid = ip->ino_rec.base.base.obj_id; 398 vap->va_mode = ip->ino_data.mode; 399 vap->va_nlink = ip->ino_rec.ino_nlinks; 400 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid); 401 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid); 402 vap->va_rmajor = 0; 403 vap->va_rminor = 0; 404 vap->va_size = ip->ino_rec.ino_size; 405 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime); 406 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime); 407 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime); 408 vap->va_flags = ip->ino_data.uflags; 409 vap->va_gen = 1; /* hammer inums are unique for all time */ 410 vap->va_blocksize = 32768; /* XXX - extract from root volume */ 411 vap->va_bytes = ip->ino_rec.ino_size; 412 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type); 413 vap->va_filerev = 0; /* XXX */ 414 /* mtime uniquely identifies any adjustments made to the file */ 415 vap->va_fsmid = ip->ino_rec.ino_mtime; 416 vap->va_uid_uuid = ip->ino_data.uid; 417 vap->va_gid_uuid = ip->ino_data.gid; 418 vap->va_fsid_uuid = ip->hmp->fsid; 419 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID | 420 VA_FSID_UUID_VALID; 421 return(0); 422 } 423 424 /* 425 * hammer_vop_nresolve { nch, dvp, cred } 426 * 427 * Locate the requested directory entry. 428 */ 429 static 430 int 431 hammer_vop_nresolve(struct vop_nresolve_args *ap) 432 { 433 struct namecache *ncp; 434 struct hammer_inode *dip; 435 struct hammer_cursor cursor; 436 union hammer_record_ondisk *rec; 437 struct vnode *vp; 438 int64_t namekey; 439 int error; 440 441 /* 442 * Calculate the namekey and setup the key range for the scan. This 443 * works kinda like a chained hash table where the lower 32 bits 444 * of the namekey synthesize the chain. 445 * 446 * The key range is inclusive of both key_beg and key_end. 447 */ 448 dip = VTOI(ap->a_dvp); 449 ncp = ap->a_nch->ncp; 450 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen); 451 452 hammer_init_cursor_ip(&cursor, dip); 453 cursor.key_beg.obj_id = dip->obj_id; 454 cursor.key_beg.key = namekey; 455 cursor.key_beg.create_tid = dip->obj_asof; 456 cursor.key_beg.delete_tid = 0; 457 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY; 458 cursor.key_beg.obj_type = 0; 459 460 cursor.key_end = cursor.key_beg; 461 cursor.key_end.key |= 0xFFFFFFFFULL; 462 463 /* 464 * Scan all matching records (the chain), locate the one matching 465 * the requested path component. info->last_error contains the 466 * error code on search termination and could be 0, ENOENT, or 467 * something else. 468 * 469 * The hammer_ip_*() functions merge in-memory records with on-disk 470 * records for the purposes of the search. 471 */ 472 rec = hammer_ip_first(&cursor, dip); 473 while (rec) { 474 if (hammer_ip_resolve_data(&cursor) != 0) /* sets last_error */ 475 break; 476 if (ncp->nc_nlen == rec->entry.base.data_len && 477 bcmp(ncp->nc_name, (void *)cursor.data, ncp->nc_nlen) == 0) { 478 break; 479 } 480 rec = hammer_ip_next(&cursor); 481 } 482 error = cursor.last_error; 483 if (error == 0) { 484 error = hammer_vfs_vget(dip->hmp->mp, rec->entry.obj_id, &vp); 485 if (error == 0) { 486 vn_unlock(vp); 487 cache_setvp(ap->a_nch, vp); 488 vrele(vp); 489 } 490 } else if (error == ENOENT) { 491 cache_setvp(ap->a_nch, NULL); 492 } 493 hammer_done_cursor(&cursor); 494 return (error); 495 } 496 497 /* 498 * hammer_vop_nlookupdotdot { dvp, vpp, cred } 499 * 500 * Locate the parent directory of a directory vnode. 501 * 502 * dvp is referenced but not locked. *vpp must be returned referenced and 503 * locked. A parent_obj_id of 0 does not necessarily indicate that we are 504 * at the root, instead it could indicate that the directory we were in was 505 * removed. 506 */ 507 static 508 int 509 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 510 { 511 struct hammer_inode *dip; 512 u_int64_t parent_obj_id; 513 514 dip = VTOI(ap->a_dvp); 515 if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) { 516 *ap->a_vpp = NULL; 517 return ENOENT; 518 } 519 return(hammer_vfs_vget(dip->hmp->mp, parent_obj_id, ap->a_vpp)); 520 } 521 522 /* 523 * hammer_vop_nlink { nch, dvp, vp, cred } 524 */ 525 static 526 int 527 hammer_vop_nlink(struct vop_nlink_args *ap) 528 { 529 struct hammer_transaction trans; 530 struct hammer_inode *dip; 531 struct hammer_inode *ip; 532 struct nchandle *nch; 533 int error; 534 535 nch = ap->a_nch; 536 dip = VTOI(ap->a_dvp); 537 ip = VTOI(ap->a_vp); 538 539 /* 540 * Create a transaction to cover the operations we perform. 541 */ 542 hammer_start_transaction(&trans, dip->hmp); 543 544 /* 545 * Add the filesystem object to the directory. Note that neither 546 * dip nor ip are referenced or locked, but their vnodes are 547 * referenced. This function will bump the inode's link count. 548 */ 549 error = hammer_add_directory(&trans, dip, nch->ncp, ip); 550 551 /* 552 * Finish up. 553 */ 554 if (error) { 555 hammer_abort_transaction(&trans); 556 } else { 557 hammer_commit_transaction(&trans); 558 } 559 return (error); 560 } 561 562 /* 563 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap } 564 * 565 * The operating system has already ensured that the directory entry 566 * does not exist and done all appropriate namespace locking. 567 */ 568 static 569 int 570 hammer_vop_nmkdir(struct vop_nmkdir_args *ap) 571 { 572 struct hammer_transaction trans; 573 struct hammer_inode *dip; 574 struct hammer_inode *nip; 575 struct nchandle *nch; 576 int error; 577 578 nch = ap->a_nch; 579 dip = VTOI(ap->a_dvp); 580 581 /* 582 * Create a transaction to cover the operations we perform. 583 */ 584 hammer_start_transaction(&trans, dip->hmp); 585 586 /* 587 * Create a new filesystem object of the requested type. The 588 * returned inode will be referenced but not locked. 589 */ 590 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip); 591 if (error) { 592 hammer_abort_transaction(&trans); 593 *ap->a_vpp = NULL; 594 return (error); 595 } 596 597 /* 598 * Add the new filesystem object to the directory. This will also 599 * bump the inode's link count. 600 */ 601 error = hammer_add_directory(&trans, dip, nch->ncp, nip); 602 603 /* 604 * Finish up. 605 */ 606 if (error) { 607 hammer_rel_inode(nip); 608 hammer_abort_transaction(&trans); 609 *ap->a_vpp = NULL; 610 } else { 611 hammer_commit_transaction(&trans); 612 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp); 613 hammer_rel_inode(nip); 614 } 615 return (error); 616 } 617 618 /* 619 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap } 620 * 621 * The operating system has already ensured that the directory entry 622 * does not exist and done all appropriate namespace locking. 623 */ 624 static 625 int 626 hammer_vop_nmknod(struct vop_nmknod_args *ap) 627 { 628 struct hammer_transaction trans; 629 struct hammer_inode *dip; 630 struct hammer_inode *nip; 631 struct nchandle *nch; 632 int error; 633 634 nch = ap->a_nch; 635 dip = VTOI(ap->a_dvp); 636 637 /* 638 * Create a transaction to cover the operations we perform. 639 */ 640 hammer_start_transaction(&trans, dip->hmp); 641 642 /* 643 * Create a new filesystem object of the requested type. The 644 * returned inode will be referenced but not locked. 645 */ 646 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip); 647 if (error) { 648 hammer_abort_transaction(&trans); 649 *ap->a_vpp = NULL; 650 return (error); 651 } 652 653 /* 654 * Add the new filesystem object to the directory. This will also 655 * bump the inode's link count. 656 */ 657 error = hammer_add_directory(&trans, dip, nch->ncp, nip); 658 659 /* 660 * Finish up. 661 */ 662 if (error) { 663 hammer_rel_inode(nip); 664 hammer_abort_transaction(&trans); 665 *ap->a_vpp = NULL; 666 } else { 667 hammer_commit_transaction(&trans); 668 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp); 669 hammer_rel_inode(nip); 670 } 671 return (error); 672 } 673 674 /* 675 * hammer_vop_open { vp, mode, cred, fp } 676 */ 677 static 678 int 679 hammer_vop_open(struct vop_open_args *ap) 680 { 681 return EOPNOTSUPP; 682 } 683 684 /* 685 * hammer_vop_pathconf { vp, name, retval } 686 */ 687 static 688 int 689 hammer_vop_pathconf(struct vop_pathconf_args *ap) 690 { 691 return EOPNOTSUPP; 692 } 693 694 /* 695 * hammer_vop_print { vp } 696 */ 697 static 698 int 699 hammer_vop_print(struct vop_print_args *ap) 700 { 701 return EOPNOTSUPP; 702 } 703 704 /* 705 * hammer_vop_readdir { vp, uio, cred, *eofflag } 706 */ 707 static 708 int 709 hammer_vop_readdir(struct vop_readdir_args *ap) 710 { 711 return EOPNOTSUPP; 712 } 713 714 /* 715 * hammer_vop_readlink { vp, uio, cred } 716 */ 717 static 718 int 719 hammer_vop_readlink(struct vop_readlink_args *ap) 720 { 721 return EOPNOTSUPP; 722 } 723 724 /* 725 * hammer_vop_nremove { nch, dvp, cred } 726 */ 727 static 728 int 729 hammer_vop_nremove(struct vop_nremove_args *ap) 730 { 731 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0)); 732 } 733 734 /* 735 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred } 736 */ 737 static 738 int 739 hammer_vop_nrename(struct vop_nrename_args *ap) 740 { 741 struct hammer_transaction trans; 742 struct namecache *fncp; 743 struct namecache *tncp; 744 struct hammer_inode *fdip; 745 struct hammer_inode *tdip; 746 struct hammer_inode *ip; 747 struct hammer_cursor cursor; 748 union hammer_record_ondisk *rec; 749 int64_t namekey; 750 int error; 751 752 fdip = VTOI(ap->a_fdvp); 753 tdip = VTOI(ap->a_tdvp); 754 fncp = ap->a_fnch->ncp; 755 tncp = ap->a_tnch->ncp; 756 hammer_start_transaction(&trans, fdip->hmp); 757 758 /* 759 * Extract the hammer_inode from fncp and add link to the target 760 * directory. 761 */ 762 ip = VTOI(fncp->nc_vp); 763 KKASSERT(ip != NULL); 764 765 error = hammer_add_directory(&trans, tdip, tncp, ip); 766 767 /* 768 * Locate the record in the originating directory and remove it. 769 * 770 * Calculate the namekey and setup the key range for the scan. This 771 * works kinda like a chained hash table where the lower 32 bits 772 * of the namekey synthesize the chain. 773 * 774 * The key range is inclusive of both key_beg and key_end. 775 */ 776 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen); 777 778 hammer_init_cursor_ip(&cursor, fdip); 779 cursor.key_beg.obj_id = fdip->obj_id; 780 cursor.key_beg.key = namekey; 781 cursor.key_beg.create_tid = fdip->obj_asof; 782 cursor.key_beg.delete_tid = 0; 783 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY; 784 cursor.key_beg.obj_type = 0; 785 786 cursor.key_end = cursor.key_beg; 787 cursor.key_end.key |= 0xFFFFFFFFULL; 788 789 /* 790 * Scan all matching records (the chain), locate the one matching 791 * the requested path component. info->last_error contains the 792 * error code on search termination and could be 0, ENOENT, or 793 * something else. 794 * 795 * The hammer_ip_*() functions merge in-memory records with on-disk 796 * records for the purposes of the search. 797 */ 798 rec = hammer_ip_first(&cursor, fdip); 799 while (rec) { 800 if (hammer_ip_resolve_data(&cursor) != 0) 801 break; 802 if (fncp->nc_nlen == rec->entry.base.data_len && 803 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) { 804 break; 805 } 806 rec = hammer_ip_next(&cursor); 807 } 808 error = cursor.last_error; 809 810 /* 811 * If all is ok we have to get the inode so we can adjust nlinks. 812 */ 813 if (error) 814 goto done; 815 error = hammer_del_directory(&trans, &cursor, fdip, ip); 816 if (error == 0) { 817 cache_rename(ap->a_fnch, ap->a_tnch); 818 cache_setvp(ap->a_tnch, ip->vp); 819 } 820 done: 821 if (error == 0) { 822 hammer_commit_transaction(&trans); 823 } else { 824 hammer_abort_transaction(&trans); 825 } 826 hammer_done_cursor(&cursor); 827 return (error); 828 } 829 830 /* 831 * hammer_vop_nrmdir { nch, dvp, cred } 832 */ 833 static 834 int 835 hammer_vop_nrmdir(struct vop_nrmdir_args *ap) 836 { 837 /* XXX check that directory is empty */ 838 839 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0)); 840 } 841 842 /* 843 * hammer_vop_setattr { vp, vap, cred } 844 */ 845 static 846 int 847 hammer_vop_setattr(struct vop_setattr_args *ap) 848 { 849 struct hammer_transaction trans; 850 struct vattr *vap; 851 struct hammer_inode *ip; 852 int modflags; 853 int error; 854 u_int32_t flags; 855 uuid_t uuid; 856 857 vap = ap->a_vap; 858 ip = ap->a_vp->v_data; 859 modflags = 0; 860 861 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) 862 return(EROFS); 863 864 hammer_start_transaction(&trans, ip->hmp); 865 error = 0; 866 867 if (vap->va_flags != VNOVAL) { 868 flags = ip->ino_data.uflags; 869 error = vop_helper_setattr_flags(&flags, vap->va_flags, 870 hammer_to_unix_xid(&ip->ino_data.uid), 871 ap->a_cred); 872 if (error == 0) { 873 if (ip->ino_data.uflags != flags) { 874 ip->ino_data.uflags = flags; 875 modflags |= HAMMER_INODE_DDIRTY; 876 } 877 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) { 878 error = 0; 879 goto done; 880 } 881 } 882 goto done; 883 } 884 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) { 885 error = EPERM; 886 goto done; 887 } 888 if (vap->va_uid != (uid_t)VNOVAL) { 889 hammer_guid_to_uuid(&uuid, vap->va_uid); 890 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) == 0) { 891 ip->ino_data.uid = uuid; 892 modflags |= HAMMER_INODE_DDIRTY; 893 } 894 } 895 if (vap->va_gid != (uid_t)VNOVAL) { 896 hammer_guid_to_uuid(&uuid, vap->va_uid); 897 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) == 0) { 898 ip->ino_data.gid = uuid; 899 modflags |= HAMMER_INODE_DDIRTY; 900 } 901 } 902 if (vap->va_size != VNOVAL) { 903 switch(ap->a_vp->v_type) { 904 case VREG: 905 case VDATABASE: 906 error = hammer_delete_range(&trans, ip, 907 vap->va_size, 908 0x7FFFFFFFFFFFFFFFLL); 909 break; 910 default: 911 error = EINVAL; 912 goto done; 913 } 914 } 915 if (vap->va_atime.tv_sec != VNOVAL) { 916 ip->ino_rec.ino_atime = 917 hammer_timespec_to_transid(&vap->va_atime); 918 modflags |= HAMMER_INODE_ITIMES; 919 } 920 if (vap->va_mtime.tv_sec != VNOVAL) { 921 ip->ino_rec.ino_mtime = 922 hammer_timespec_to_transid(&vap->va_mtime); 923 modflags |= HAMMER_INODE_ITIMES; 924 } 925 if (vap->va_mode != (mode_t)VNOVAL) { 926 if (ip->ino_data.mode != vap->va_mode) { 927 ip->ino_data.mode = vap->va_mode; 928 modflags |= HAMMER_INODE_DDIRTY; 929 } 930 } 931 done: 932 if (error) { 933 hammer_abort_transaction(&trans); 934 } else { 935 if (modflags) 936 hammer_modify_inode(&trans, ip, modflags); 937 hammer_commit_transaction(&trans); 938 } 939 return (error); 940 } 941 942 /* 943 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target } 944 */ 945 static 946 int 947 hammer_vop_nsymlink(struct vop_nsymlink_args *ap) 948 { 949 return EOPNOTSUPP; 950 } 951 952 /* 953 * hammer_vop_nwhiteout { nch, dvp, cred, flags } 954 */ 955 static 956 int 957 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap) 958 { 959 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags)); 960 } 961 962 /* 963 * hammer_vop_strategy { vp, bio } 964 * 965 * Strategy call, used for regular file read & write only. Note that the 966 * bp may represent a cluster. 967 * 968 * To simplify operation and allow better optimizations in the future, 969 * this code does not make any assumptions with regards to buffer alignment 970 * or size. 971 */ 972 static 973 int 974 hammer_vop_strategy(struct vop_strategy_args *ap) 975 { 976 struct buf *bp; 977 int error; 978 979 bp = ap->a_bio->bio_buf; 980 981 switch(bp->b_cmd) { 982 case BUF_CMD_READ: 983 error = hammer_vop_strategy_read(ap); 984 break; 985 case BUF_CMD_WRITE: 986 error = hammer_vop_strategy_write(ap); 987 break; 988 default: 989 error = EINVAL; 990 break; 991 } 992 bp->b_error = error; 993 if (error) 994 bp->b_flags |= B_ERROR; 995 biodone(ap->a_bio); 996 return (error); 997 } 998 999 /* 1000 * Read from a regular file. Iterate the related records and fill in the 1001 * BIO/BUF. Gaps are zero-filled. 1002 * 1003 * The support code in hammer_object.c should be used to deal with mixed 1004 * in-memory and on-disk records. 1005 * 1006 * XXX atime update 1007 */ 1008 static 1009 int 1010 hammer_vop_strategy_read(struct vop_strategy_args *ap) 1011 { 1012 struct hammer_inode *ip = ap->a_vp->v_data; 1013 struct hammer_cursor cursor; 1014 hammer_record_ondisk_t rec; 1015 hammer_base_elm_t base; 1016 struct bio *bio; 1017 struct buf *bp; 1018 int64_t rec_offset; 1019 int error; 1020 int boff; 1021 int roff; 1022 int n; 1023 1024 bio = ap->a_bio; 1025 bp = bio->bio_buf; 1026 1027 hammer_init_cursor_ip(&cursor, ip); 1028 1029 /* 1030 * Key range (begin and end inclusive) to scan. Note that the key's 1031 * stored in the actual records represent the 1032 */ 1033 cursor.key_beg.obj_id = ip->obj_id; 1034 cursor.key_beg.create_tid = ip->obj_asof; 1035 cursor.key_beg.delete_tid = 0; 1036 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA; 1037 cursor.key_beg.obj_type = 0; 1038 cursor.key_beg.key = bio->bio_offset; 1039 1040 cursor.key_end = cursor.key_beg; 1041 cursor.key_end.key = bio->bio_offset + bp->b_bufsize - 1; 1042 1043 rec = hammer_ip_first(&cursor, ip); 1044 boff = 0; 1045 1046 while (rec) { 1047 if (hammer_ip_resolve_data(&cursor) != 0) 1048 break; 1049 base = &rec->base.base; 1050 1051 rec_offset = base->key - rec->data.base.data_len; 1052 1053 /* 1054 * Zero-fill any gap 1055 */ 1056 n = (int)(rec_offset - (bio->bio_offset + boff)); 1057 if (n > 0) { 1058 kprintf("zfill %d bytes\n", n); 1059 bzero((char *)bp->b_data + boff, n); 1060 boff += n; 1061 n = 0; 1062 } 1063 1064 /* 1065 * Calculate the data offset in the record and the number 1066 * of bytes we can copy. 1067 */ 1068 roff = -n; 1069 n = rec->data.base.data_len - roff; 1070 KKASSERT(n > 0); 1071 if (n > bp->b_bufsize - boff) 1072 n = bp->b_bufsize - boff; 1073 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n); 1074 boff += n; 1075 if (boff == bp->b_bufsize) 1076 break; 1077 rec = hammer_ip_next(&cursor); 1078 } 1079 hammer_done_cursor(&cursor); 1080 1081 /* 1082 * There may have been a gap after the last record 1083 */ 1084 error = cursor.last_error; 1085 if (error == ENOENT) 1086 error = 0; 1087 if (error == 0 && boff != bp->b_bufsize) { 1088 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff); 1089 /* boff = bp->b_bufsize; */ 1090 } 1091 bp->b_resid = 0; 1092 return(error); 1093 } 1094 1095 /* 1096 * Write to a regular file. Iterate the related records and mark for 1097 * deletion. If existing edge records (left and right side) overlap our 1098 * write they have to be marked deleted and new records created, usually 1099 * referencing a portion of the original data. Then add a record to 1100 * represent the buffer. 1101 * 1102 * The support code in hammer_object.c should be used to deal with mixed 1103 * in-memory and on-disk records. 1104 */ 1105 static 1106 int 1107 hammer_vop_strategy_write(struct vop_strategy_args *ap) 1108 { 1109 struct hammer_transaction trans; 1110 hammer_inode_t ip; 1111 struct bio *bio; 1112 struct buf *bp; 1113 int error; 1114 1115 bio = ap->a_bio; 1116 bp = bio->bio_buf; 1117 ip = ap->a_vp->v_data; 1118 hammer_start_transaction(&trans, ip->hmp); 1119 1120 /* 1121 * Delete any records overlapping our range. This function will 1122 * properly 1123 */ 1124 error = hammer_delete_range(&trans, ip, bio->bio_offset, 1125 bio->bio_offset + bp->b_bufsize - 1); 1126 1127 /* 1128 * Add a single record to cover the write 1129 */ 1130 if (error == 0) { 1131 error = hammer_add_data(&trans, ip, bio->bio_offset, 1132 bp->b_data, bp->b_bufsize); 1133 } 1134 1135 /* 1136 * If an error occured abort the transaction 1137 */ 1138 if (error) { 1139 /* XXX undo deletion */ 1140 hammer_abort_transaction(&trans); 1141 bp->b_resid = bp->b_bufsize; 1142 } else { 1143 hammer_commit_transaction(&trans); 1144 bp->b_resid = 0; 1145 } 1146 return(error); 1147 } 1148 1149 /* 1150 * dounlink - disconnect a directory entry 1151 * 1152 * XXX whiteout support not really in yet 1153 */ 1154 static int 1155 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred, 1156 int flags) 1157 { 1158 struct hammer_transaction trans; 1159 struct namecache *ncp; 1160 hammer_inode_t dip; 1161 hammer_inode_t ip; 1162 hammer_record_ondisk_t rec; 1163 struct hammer_cursor cursor; 1164 struct vnode *vp; 1165 int64_t namekey; 1166 int error; 1167 1168 /* 1169 * Calculate the namekey and setup the key range for the scan. This 1170 * works kinda like a chained hash table where the lower 32 bits 1171 * of the namekey synthesize the chain. 1172 * 1173 * The key range is inclusive of both key_beg and key_end. 1174 */ 1175 dip = VTOI(dvp); 1176 ncp = nch->ncp; 1177 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen); 1178 1179 hammer_init_cursor_ip(&cursor, dip); 1180 cursor.key_beg.obj_id = dip->obj_id; 1181 cursor.key_beg.key = namekey; 1182 cursor.key_beg.create_tid = dip->obj_asof; 1183 cursor.key_beg.delete_tid = 0; 1184 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY; 1185 cursor.key_beg.obj_type = 0; 1186 1187 cursor.key_end = cursor.key_beg; 1188 cursor.key_end.key |= 0xFFFFFFFFULL; 1189 1190 hammer_start_transaction(&trans, dip->hmp); 1191 1192 /* 1193 * Scan all matching records (the chain), locate the one matching 1194 * the requested path component. info->last_error contains the 1195 * error code on search termination and could be 0, ENOENT, or 1196 * something else. 1197 * 1198 * The hammer_ip_*() functions merge in-memory records with on-disk 1199 * records for the purposes of the search. 1200 */ 1201 rec = hammer_ip_first(&cursor, dip); 1202 while (rec) { 1203 if (hammer_ip_resolve_data(&cursor) != 0) 1204 break; 1205 if (ncp->nc_nlen == rec->entry.base.data_len && 1206 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) { 1207 break; 1208 } 1209 rec = hammer_ip_next(&cursor); 1210 } 1211 error = cursor.last_error; 1212 1213 /* 1214 * If all is ok we have to get the inode so we can adjust nlinks. 1215 */ 1216 if (error == 0) { 1217 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id, &error); 1218 if (error == 0) 1219 error = hammer_del_directory(&trans, &cursor, dip, ip); 1220 if (error == 0) { 1221 cache_setunresolved(nch); 1222 cache_setvp(nch, NULL); 1223 /* XXX locking */ 1224 if (ip->vp) 1225 cache_inval_vp(ip->vp, CINV_DESTROY); 1226 } 1227 hammer_rel_inode(ip); 1228 1229 error = hammer_vfs_vget(dip->hmp->mp, rec->entry.obj_id, &vp); 1230 if (error == 0) { 1231 vn_unlock(vp); 1232 cache_setvp(nch, vp); 1233 vrele(vp); 1234 hammer_commit_transaction(&trans); 1235 } else { 1236 hammer_abort_transaction(&trans); 1237 } 1238 } 1239 hammer_done_cursor(&cursor); 1240 return (error); 1241 } 1242 1243