1 /* 2 * Copyright (c) 2004-2006 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 /* 35 * Each mount point may have zero or more independantly configured journals 36 * attached to it. Each journal is represented by a memory FIFO and worker 37 * thread. Journal events are streamed through the FIFO to the thread, 38 * batched up (typically on one-second intervals), and written out by the 39 * thread. 40 * 41 * Journal vnode ops are executed instead of mnt_vn_norm_ops when one or 42 * more journals have been installed on a mount point. It becomes the 43 * responsibility of the journal op to call the underlying normal op as 44 * appropriate. 45 */ 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/buf.h> 49 #include <sys/conf.h> 50 #include <sys/kernel.h> 51 #include <sys/queue.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mount.h> 55 #include <sys/unistd.h> 56 #include <sys/vnode.h> 57 #include <sys/poll.h> 58 #include <sys/mountctl.h> 59 #include <sys/journal.h> 60 #include <sys/file.h> 61 #include <sys/proc.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 65 #include <machine/limits.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_object.h> 69 #include <vm/vm_page.h> 70 #include <vm/vm_pager.h> 71 #include <vm/vnode_pager.h> 72 73 #include <sys/file2.h> 74 75 static int journal_attach(struct mount *mp); 76 static void journal_detach(struct mount *mp); 77 static int journal_install_vfs_journal(struct mount *mp, struct file *fp, 78 const struct mountctl_install_journal *info); 79 static int journal_restart_vfs_journal(struct mount *mp, struct file *fp, 80 const struct mountctl_restart_journal *info); 81 static int journal_remove_vfs_journal(struct mount *mp, 82 const struct mountctl_remove_journal *info); 83 static int journal_restart(struct mount *mp, struct file *fp, 84 struct journal *jo, int flags); 85 static int journal_destroy(struct mount *mp, struct journal *jo, int flags); 86 static int journal_resync_vfs_journal(struct mount *mp, const void *ctl); 87 static int journal_status_vfs_journal(struct mount *mp, 88 const struct mountctl_status_journal *info, 89 struct mountctl_journal_ret_status *rstat, 90 int buflen, int *res); 91 92 static void jrecord_undo_file(struct jrecord *jrec, struct vnode *vp, 93 int jrflags, off_t off, off_t bytes); 94 95 static int journal_setattr(struct vop_setattr_args *ap); 96 static int journal_write(struct vop_write_args *ap); 97 static int journal_fsync(struct vop_fsync_args *ap); 98 static int journal_putpages(struct vop_putpages_args *ap); 99 static int journal_setacl(struct vop_setacl_args *ap); 100 static int journal_setextattr(struct vop_setextattr_args *ap); 101 static int journal_ncreate(struct vop_ncreate_args *ap); 102 static int journal_nmknod(struct vop_nmknod_args *ap); 103 static int journal_nlink(struct vop_nlink_args *ap); 104 static int journal_nsymlink(struct vop_nsymlink_args *ap); 105 static int journal_nwhiteout(struct vop_nwhiteout_args *ap); 106 static int journal_nremove(struct vop_nremove_args *ap); 107 static int journal_nmkdir(struct vop_nmkdir_args *ap); 108 static int journal_nrmdir(struct vop_nrmdir_args *ap); 109 static int journal_nrename(struct vop_nrename_args *ap); 110 111 #define JRUNDO_SIZE 0x00000001 112 #define JRUNDO_UID 0x00000002 113 #define JRUNDO_GID 0x00000004 114 #define JRUNDO_FSID 0x00000008 115 #define JRUNDO_MODES 0x00000010 116 #define JRUNDO_INUM 0x00000020 117 #define JRUNDO_ATIME 0x00000040 118 #define JRUNDO_MTIME 0x00000080 119 #define JRUNDO_CTIME 0x00000100 120 #define JRUNDO_GEN 0x00000200 121 #define JRUNDO_FLAGS 0x00000400 122 #define JRUNDO_UDEV 0x00000800 123 #define JRUNDO_NLINK 0x00001000 124 #define JRUNDO_FILEDATA 0x00010000 125 #define JRUNDO_GETVP 0x00020000 126 #define JRUNDO_CONDLINK 0x00040000 /* write file data if link count 1 */ 127 #define JRUNDO_VATTR (JRUNDO_SIZE|JRUNDO_UID|JRUNDO_GID|JRUNDO_FSID|\ 128 JRUNDO_MODES|JRUNDO_INUM|JRUNDO_ATIME|JRUNDO_MTIME|\ 129 JRUNDO_CTIME|JRUNDO_GEN|JRUNDO_FLAGS|JRUNDO_UDEV|\ 130 JRUNDO_NLINK) 131 #define JRUNDO_ALL (JRUNDO_VATTR|JRUNDO_FILEDATA) 132 133 static struct vop_ops journal_vnode_vops = { 134 .vop_default = vop_journal_operate_ap, 135 .vop_mountctl = journal_mountctl, 136 .vop_setattr = journal_setattr, 137 .vop_write = journal_write, 138 .vop_fsync = journal_fsync, 139 .vop_putpages = journal_putpages, 140 .vop_setacl = journal_setacl, 141 .vop_setextattr = journal_setextattr, 142 .vop_ncreate = journal_ncreate, 143 .vop_nmknod = journal_nmknod, 144 .vop_nlink = journal_nlink, 145 .vop_nsymlink = journal_nsymlink, 146 .vop_nwhiteout = journal_nwhiteout, 147 .vop_nremove = journal_nremove, 148 .vop_nmkdir = journal_nmkdir, 149 .vop_nrmdir = journal_nrmdir, 150 .vop_nrename = journal_nrename 151 }; 152 153 int 154 journal_mountctl(struct vop_mountctl_args *ap) 155 { 156 struct mount *mp; 157 int error = 0; 158 159 mp = ap->a_head.a_ops->head.vv_mount; 160 KKASSERT(mp); 161 162 if (mp->mnt_vn_journal_ops == NULL) { 163 switch(ap->a_op) { 164 case MOUNTCTL_INSTALL_VFS_JOURNAL: 165 error = journal_attach(mp); 166 if (error == 0 && ap->a_ctllen != sizeof(struct mountctl_install_journal)) 167 error = EINVAL; 168 if (error == 0 && ap->a_fp == NULL) 169 error = EBADF; 170 if (error == 0) 171 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl); 172 if (TAILQ_EMPTY(&mp->mnt_jlist)) 173 journal_detach(mp); 174 break; 175 case MOUNTCTL_RESTART_VFS_JOURNAL: 176 case MOUNTCTL_REMOVE_VFS_JOURNAL: 177 case MOUNTCTL_RESYNC_VFS_JOURNAL: 178 case MOUNTCTL_STATUS_VFS_JOURNAL: 179 error = ENOENT; 180 break; 181 default: 182 error = EOPNOTSUPP; 183 break; 184 } 185 } else { 186 switch(ap->a_op) { 187 case MOUNTCTL_INSTALL_VFS_JOURNAL: 188 if (ap->a_ctllen != sizeof(struct mountctl_install_journal)) 189 error = EINVAL; 190 if (error == 0 && ap->a_fp == NULL) 191 error = EBADF; 192 if (error == 0) 193 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl); 194 break; 195 case MOUNTCTL_RESTART_VFS_JOURNAL: 196 if (ap->a_ctllen != sizeof(struct mountctl_restart_journal)) 197 error = EINVAL; 198 if (error == 0 && ap->a_fp == NULL) 199 error = EBADF; 200 if (error == 0) 201 error = journal_restart_vfs_journal(mp, ap->a_fp, ap->a_ctl); 202 break; 203 case MOUNTCTL_REMOVE_VFS_JOURNAL: 204 if (ap->a_ctllen != sizeof(struct mountctl_remove_journal)) 205 error = EINVAL; 206 if (error == 0) 207 error = journal_remove_vfs_journal(mp, ap->a_ctl); 208 if (TAILQ_EMPTY(&mp->mnt_jlist)) 209 journal_detach(mp); 210 break; 211 case MOUNTCTL_RESYNC_VFS_JOURNAL: 212 if (ap->a_ctllen != 0) 213 error = EINVAL; 214 error = journal_resync_vfs_journal(mp, ap->a_ctl); 215 break; 216 case MOUNTCTL_STATUS_VFS_JOURNAL: 217 if (ap->a_ctllen != sizeof(struct mountctl_status_journal)) 218 error = EINVAL; 219 if (error == 0) { 220 error = journal_status_vfs_journal(mp, ap->a_ctl, 221 ap->a_buf, ap->a_buflen, ap->a_res); 222 } 223 break; 224 default: 225 error = EOPNOTSUPP; 226 break; 227 } 228 } 229 return (error); 230 } 231 232 /* 233 * High level mount point setup. When a 234 */ 235 static int 236 journal_attach(struct mount *mp) 237 { 238 KKASSERT(mp->mnt_jbitmap == NULL); 239 vfs_add_vnodeops(mp, &journal_vnode_vops, &mp->mnt_vn_journal_ops); 240 mp->mnt_jbitmap = kmalloc(JREC_STREAMID_JMAX/8, M_JOURNAL, M_WAITOK|M_ZERO); 241 mp->mnt_streamid = JREC_STREAMID_JMIN; 242 return(0); 243 } 244 245 static void 246 journal_detach(struct mount *mp) 247 { 248 KKASSERT(mp->mnt_jbitmap != NULL); 249 if (mp->mnt_vn_journal_ops) 250 vfs_rm_vnodeops(mp, &journal_vnode_vops, &mp->mnt_vn_journal_ops); 251 kfree(mp->mnt_jbitmap, M_JOURNAL); 252 mp->mnt_jbitmap = NULL; 253 } 254 255 /* 256 * Install a journal on a mount point. Each journal has an associated worker 257 * thread which is responsible for buffering and spooling the data to the 258 * target. A mount point may have multiple journals attached to it. An 259 * initial start record is generated when the journal is associated. 260 */ 261 static int 262 journal_install_vfs_journal(struct mount *mp, struct file *fp, 263 const struct mountctl_install_journal *info) 264 { 265 struct journal *jo; 266 struct jrecord jrec; 267 int error = 0; 268 int size; 269 270 jo = kmalloc(sizeof(struct journal), M_JOURNAL, M_WAITOK|M_ZERO); 271 bcopy(info->id, jo->id, sizeof(jo->id)); 272 jo->flags = info->flags & ~(MC_JOURNAL_WACTIVE | MC_JOURNAL_RACTIVE | 273 MC_JOURNAL_STOP_REQ); 274 275 /* 276 * Memory FIFO size, round to nearest power of 2 277 */ 278 if (info->membufsize) { 279 if (info->membufsize < 65536) 280 size = 65536; 281 else if (info->membufsize > 128 * 1024 * 1024) 282 size = 128 * 1024 * 1024; 283 else 284 size = (int)info->membufsize; 285 } else { 286 size = 1024 * 1024; 287 } 288 jo->fifo.size = 1; 289 while (jo->fifo.size < size) 290 jo->fifo.size <<= 1; 291 292 /* 293 * Other parameters. If not specified the starting transaction id 294 * will be the current date. 295 */ 296 if (info->transid) { 297 jo->transid = info->transid; 298 } else { 299 struct timespec ts; 300 getnanotime(&ts); 301 jo->transid = ((int64_t)ts.tv_sec << 30) | ts.tv_nsec; 302 } 303 304 jo->fp = fp; 305 306 /* 307 * Allocate the memory FIFO 308 */ 309 jo->fifo.mask = jo->fifo.size - 1; 310 jo->fifo.membase = kmalloc(jo->fifo.size, M_JFIFO, M_WAITOK|M_ZERO|M_NULLOK); 311 if (jo->fifo.membase == NULL) 312 error = ENOMEM; 313 314 /* 315 * Create the worker threads and generate the association record. 316 */ 317 if (error) { 318 kfree(jo, M_JOURNAL); 319 } else { 320 fhold(fp); 321 journal_create_threads(jo); 322 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT); 323 jrecord_write(&jrec, JTYPE_ASSOCIATE, 0); 324 jrecord_done(&jrec, 0); 325 TAILQ_INSERT_TAIL(&mp->mnt_jlist, jo, jentry); 326 } 327 return(error); 328 } 329 330 /* 331 * Restart a journal with a new descriptor. The existing reader and writer 332 * threads are terminated and a new descriptor is associated with the 333 * journal. The FIFO rindex is reset to xindex and the threads are then 334 * restarted. 335 */ 336 static int 337 journal_restart_vfs_journal(struct mount *mp, struct file *fp, 338 const struct mountctl_restart_journal *info) 339 { 340 struct journal *jo; 341 int error; 342 343 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) { 344 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0) 345 break; 346 } 347 if (jo) 348 error = journal_restart(mp, fp, jo, info->flags); 349 else 350 error = EINVAL; 351 return (error); 352 } 353 354 static int 355 journal_restart(struct mount *mp, struct file *fp, 356 struct journal *jo, int flags) 357 { 358 /* 359 * XXX lock the jo 360 */ 361 362 #if 0 363 /* 364 * Record the fact that we are doing a restart in the journal. 365 * XXX it isn't safe to do this if the journal is being restarted 366 * because it was locked up and the writer thread has already exited. 367 */ 368 jrecord_init(jo, &jrec, JREC_STREAMID_RESTART); 369 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0); 370 jrecord_done(&jrec, 0); 371 #endif 372 373 /* 374 * Stop the reader and writer threads and clean up the current 375 * descriptor. 376 */ 377 kprintf("RESTART WITH FP %p KILLING %p\n", fp, jo->fp); 378 journal_destroy_threads(jo, flags); 379 380 if (jo->fp) 381 fdrop(jo->fp); 382 383 /* 384 * Associate the new descriptor, reset the FIFO index, and recreate 385 * the threads. 386 */ 387 fhold(fp); 388 jo->fp = fp; 389 jo->fifo.rindex = jo->fifo.xindex; 390 journal_create_threads(jo); 391 392 return(0); 393 } 394 395 /* 396 * Disassociate a journal from a mount point and terminate its worker thread. 397 * A final termination record is written out before the file pointer is 398 * dropped. 399 */ 400 static int 401 journal_remove_vfs_journal(struct mount *mp, 402 const struct mountctl_remove_journal *info) 403 { 404 struct journal *jo; 405 int error; 406 407 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) { 408 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0) 409 break; 410 } 411 if (jo) 412 error = journal_destroy(mp, jo, info->flags); 413 else 414 error = EINVAL; 415 return (error); 416 } 417 418 /* 419 * Remove all journals associated with a mount point. Usually called 420 * by the umount code. 421 */ 422 void 423 journal_remove_all_journals(struct mount *mp, int flags) 424 { 425 struct journal *jo; 426 427 while ((jo = TAILQ_FIRST(&mp->mnt_jlist)) != NULL) { 428 journal_destroy(mp, jo, flags); 429 } 430 } 431 432 static int 433 journal_destroy(struct mount *mp, struct journal *jo, int flags) 434 { 435 struct jrecord jrec; 436 437 TAILQ_REMOVE(&mp->mnt_jlist, jo, jentry); 438 439 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT); 440 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0); 441 jrecord_done(&jrec, 0); 442 443 journal_destroy_threads(jo, flags); 444 445 if (jo->fp) 446 fdrop(jo->fp); 447 if (jo->fifo.membase) 448 kfree(jo->fifo.membase, M_JFIFO); 449 kfree(jo, M_JOURNAL); 450 451 return(0); 452 } 453 454 static int 455 journal_resync_vfs_journal(struct mount *mp, const void *ctl) 456 { 457 return(EINVAL); 458 } 459 460 static int 461 journal_status_vfs_journal(struct mount *mp, 462 const struct mountctl_status_journal *info, 463 struct mountctl_journal_ret_status *rstat, 464 int buflen, int *res) 465 { 466 struct journal *jo; 467 int error = 0; 468 int index; 469 470 index = 0; 471 *res = 0; 472 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) { 473 if (info->index == MC_JOURNAL_INDEX_ID) { 474 if (bcmp(jo->id, info->id, sizeof(jo->id)) != 0) 475 continue; 476 } else if (info->index >= 0) { 477 if (info->index < index) 478 continue; 479 } else if (info->index != MC_JOURNAL_INDEX_ALL) { 480 continue; 481 } 482 if (buflen < sizeof(*rstat)) { 483 if (*res) 484 rstat[-1].flags |= MC_JOURNAL_STATUS_MORETOCOME; 485 else 486 error = EINVAL; 487 break; 488 } 489 bzero(rstat, sizeof(*rstat)); 490 rstat->recsize = sizeof(*rstat); 491 bcopy(jo->id, rstat->id, sizeof(jo->id)); 492 rstat->index = index; 493 rstat->membufsize = jo->fifo.size; 494 rstat->membufused = jo->fifo.windex - jo->fifo.xindex; 495 rstat->membufunacked = jo->fifo.rindex - jo->fifo.xindex; 496 rstat->bytessent = jo->total_acked; 497 rstat->fifostalls = jo->fifostalls; 498 ++rstat; 499 ++index; 500 *res += sizeof(*rstat); 501 buflen -= sizeof(*rstat); 502 } 503 return(error); 504 } 505 506 /************************************************************************ 507 * PARALLEL TRANSACTION SUPPORT ROUTINES * 508 ************************************************************************ 509 * 510 * JRECLIST_*() - routines which create and iterate over jrecord structures, 511 * because a mount point may have multiple attached journals. 512 */ 513 514 /* 515 * Initialize the passed jrecord_list and create a jrecord for each 516 * journal we need to write to. Unnecessary mallocs are avoided by 517 * using the passed jrecord structure as the first jrecord in the list. 518 * A starting transaction is pushed for each jrecord. 519 * 520 * Returns non-zero if any of the journals require undo records. 521 */ 522 static 523 int 524 jreclist_init(struct mount *mp, struct jrecord_list *jreclist, 525 struct jrecord *jreccache, int16_t rectype) 526 { 527 struct journal *jo; 528 struct jrecord *jrec; 529 int wantrev; 530 int count; 531 int16_t streamid; 532 533 TAILQ_INIT(&jreclist->list); 534 535 /* 536 * Select the stream ID to use for the transaction. We must select 537 * a stream ID that is not currently in use by some other parallel 538 * transaction. 539 * 540 * Don't bother calculating the next streamid when reassigning 541 * mnt_streamid, since parallel transactions are fairly rare. This 542 * also allows someone observing the raw records to clearly see 543 * when parallel transactions occur. 544 */ 545 streamid = mp->mnt_streamid; 546 count = 0; 547 while (mp->mnt_jbitmap[streamid >> 3] & (1 << (streamid & 7))) { 548 if (++streamid == JREC_STREAMID_JMAX) 549 streamid = JREC_STREAMID_JMIN; 550 if (++count == JREC_STREAMID_JMAX - JREC_STREAMID_JMIN) { 551 kprintf("jreclist_init: all streamid's in use! sleeping\n"); 552 tsleep(jreclist, 0, "jsidfl", hz * 10); 553 count = 0; 554 } 555 } 556 mp->mnt_jbitmap[streamid >> 3] |= 1 << (streamid & 7); 557 mp->mnt_streamid = streamid; 558 jreclist->streamid = streamid; 559 560 /* 561 * Now initialize a stream on each journal. 562 */ 563 count = 0; 564 wantrev = 0; 565 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) { 566 if (count == 0) 567 jrec = jreccache; 568 else 569 jrec = kmalloc(sizeof(*jrec), M_JOURNAL, M_WAITOK); 570 jrecord_init(jo, jrec, streamid); 571 jrec->user_save = jrecord_push(jrec, rectype); 572 TAILQ_INSERT_TAIL(&jreclist->list, jrec, user_entry); 573 if (jo->flags & MC_JOURNAL_WANT_REVERSABLE) 574 wantrev = 1; 575 ++count; 576 } 577 return(wantrev); 578 } 579 580 /* 581 * Terminate the journaled transactions started by jreclist_init(). If 582 * an error occured, the transaction records will be aborted. 583 */ 584 static 585 void 586 jreclist_done(struct mount *mp, struct jrecord_list *jreclist, int error) 587 { 588 struct jrecord *jrec; 589 int count; 590 591 /* 592 * Cleanup the jrecord state on each journal. 593 */ 594 TAILQ_FOREACH(jrec, &jreclist->list, user_entry) { 595 jrecord_pop(jrec, jrec->user_save); 596 jrecord_done(jrec, error); 597 } 598 599 /* 600 * Free allocated jrec's (the first is always supplied) 601 */ 602 count = 0; 603 while ((jrec = TAILQ_FIRST(&jreclist->list)) != NULL) { 604 TAILQ_REMOVE(&jreclist->list, jrec, user_entry); 605 if (count) 606 kfree(jrec, M_JOURNAL); 607 ++count; 608 } 609 610 /* 611 * Clear the streamid so it can be reused. 612 */ 613 mp->mnt_jbitmap[jreclist->streamid >> 3] &= ~(1 << (jreclist->streamid & 7)); 614 } 615 616 /* 617 * This procedure writes out UNDO records for available reversable 618 * journals. 619 * 620 * XXX could use improvement. There is no need to re-read the file 621 * for each journal. 622 */ 623 static 624 void 625 jreclist_undo_file(struct jrecord_list *jreclist, struct vnode *vp, 626 int jrflags, off_t off, off_t bytes) 627 { 628 struct jrecord *jrec; 629 int error; 630 631 error = 0; 632 if (jrflags & JRUNDO_GETVP) 633 error = vget(vp, LK_SHARED); 634 if (error == 0) { 635 TAILQ_FOREACH(jrec, &jreclist->list, user_entry) { 636 if (jrec->jo->flags & MC_JOURNAL_WANT_REVERSABLE) { 637 jrecord_undo_file(jrec, vp, jrflags, off, bytes); 638 } 639 } 640 } 641 if (error == 0 && jrflags & JRUNDO_GETVP) 642 vput(vp); 643 } 644 645 /************************************************************************ 646 * LOW LEVEL UNDO SUPPORT ROUTINE * 647 ************************************************************************ 648 * 649 * This function is used to support UNDO records. It will generate an 650 * appropriate record with the requested portion of the file data. Note 651 * that file data is only recorded if JRUNDO_FILEDATA is passed. If bytes 652 * is -1, it will be set to the size of the file. 653 */ 654 static void 655 jrecord_undo_file(struct jrecord *jrec, struct vnode *vp, int jrflags, 656 off_t off, off_t bytes) 657 { 658 struct vattr attr; 659 void *save1; /* warning, save pointers do not always remain valid */ 660 void *save2; 661 int error; 662 663 /* 664 * Setup. Start the UNDO record, obtain a shared lock on the vnode, 665 * and retrieve attribute info. 666 */ 667 save1 = jrecord_push(jrec, JTYPE_UNDO); 668 error = VOP_GETATTR(vp, &attr); 669 if (error) 670 goto done; 671 672 /* 673 * Generate UNDO records as requested. 674 */ 675 if (jrflags & JRUNDO_VATTR) { 676 save2 = jrecord_push(jrec, JTYPE_VATTR); 677 jrecord_leaf(jrec, JLEAF_VTYPE, &attr.va_type, sizeof(attr.va_type)); 678 if ((jrflags & JRUNDO_NLINK) && attr.va_nlink != VNOVAL) 679 jrecord_leaf(jrec, JLEAF_NLINK, &attr.va_nlink, sizeof(attr.va_nlink)); 680 if ((jrflags & JRUNDO_SIZE) && attr.va_size != VNOVAL) 681 jrecord_leaf(jrec, JLEAF_SIZE, &attr.va_size, sizeof(attr.va_size)); 682 if ((jrflags & JRUNDO_UID) && attr.va_uid != VNOVAL) 683 jrecord_leaf(jrec, JLEAF_UID, &attr.va_uid, sizeof(attr.va_uid)); 684 if ((jrflags & JRUNDO_GID) && attr.va_gid != VNOVAL) 685 jrecord_leaf(jrec, JLEAF_GID, &attr.va_gid, sizeof(attr.va_gid)); 686 if ((jrflags & JRUNDO_FSID) && attr.va_fsid != VNOVAL) 687 jrecord_leaf(jrec, JLEAF_FSID, &attr.va_fsid, sizeof(attr.va_fsid)); 688 if ((jrflags & JRUNDO_MODES) && attr.va_mode != (mode_t)VNOVAL) 689 jrecord_leaf(jrec, JLEAF_MODES, &attr.va_mode, sizeof(attr.va_mode)); 690 if ((jrflags & JRUNDO_INUM) && attr.va_fileid != VNOVAL) 691 jrecord_leaf(jrec, JLEAF_INUM, &attr.va_fileid, sizeof(attr.va_fileid)); 692 if ((jrflags & JRUNDO_ATIME) && attr.va_atime.tv_sec != VNOVAL) 693 jrecord_leaf(jrec, JLEAF_ATIME, &attr.va_atime, sizeof(attr.va_atime)); 694 if ((jrflags & JRUNDO_MTIME) && attr.va_mtime.tv_sec != VNOVAL) 695 jrecord_leaf(jrec, JLEAF_MTIME, &attr.va_mtime, sizeof(attr.va_mtime)); 696 if ((jrflags & JRUNDO_CTIME) && attr.va_ctime.tv_sec != VNOVAL) 697 jrecord_leaf(jrec, JLEAF_CTIME, &attr.va_ctime, sizeof(attr.va_ctime)); 698 if ((jrflags & JRUNDO_GEN) && attr.va_gen != VNOVAL) 699 jrecord_leaf(jrec, JLEAF_GEN, &attr.va_gen, sizeof(attr.va_gen)); 700 if ((jrflags & JRUNDO_FLAGS) && attr.va_flags != VNOVAL) 701 jrecord_leaf(jrec, JLEAF_FLAGS, &attr.va_flags, sizeof(attr.va_flags)); 702 if ((jrflags & JRUNDO_UDEV) && attr.va_rmajor != VNOVAL) { 703 udev_t rdev = makeudev(attr.va_rmajor, attr.va_rminor); 704 jrecord_leaf(jrec, JLEAF_UDEV, &rdev, sizeof(rdev)); 705 jrecord_leaf(jrec, JLEAF_UMAJOR, &attr.va_rmajor, sizeof(attr.va_rmajor)); 706 jrecord_leaf(jrec, JLEAF_UMINOR, &attr.va_rminor, sizeof(attr.va_rminor)); 707 } 708 jrecord_pop(jrec, save2); 709 } 710 711 /* 712 * Output the file data being overwritten by reading the file and 713 * writing it out to the journal prior to the write operation. We 714 * do not need to write out data past the current file EOF. 715 * 716 * XXX support JRUNDO_CONDLINK - do not write out file data for files 717 * with a link count > 1. The undo code needs to locate the inode and 718 * regenerate the hardlink. 719 */ 720 if ((jrflags & JRUNDO_FILEDATA) && attr.va_type == VREG) { 721 if (attr.va_size != VNOVAL) { 722 if (bytes == -1) 723 bytes = attr.va_size - off; 724 if (off + bytes > attr.va_size) 725 bytes = attr.va_size - off; 726 if (bytes > 0) 727 jrecord_file_data(jrec, vp, off, bytes); 728 } else { 729 error = EINVAL; 730 } 731 } 732 if ((jrflags & JRUNDO_FILEDATA) && attr.va_type == VLNK) { 733 struct iovec aiov; 734 struct uio auio; 735 char *buf; 736 737 buf = kmalloc(PATH_MAX, M_JOURNAL, M_WAITOK); 738 aiov.iov_base = buf; 739 aiov.iov_len = PATH_MAX; 740 auio.uio_iov = &aiov; 741 auio.uio_iovcnt = 1; 742 auio.uio_offset = 0; 743 auio.uio_rw = UIO_READ; 744 auio.uio_segflg = UIO_SYSSPACE; 745 auio.uio_td = curthread; 746 auio.uio_resid = PATH_MAX; 747 error = VOP_READLINK(vp, &auio, proc0.p_ucred); 748 if (error == 0) { 749 jrecord_leaf(jrec, JLEAF_SYMLINKDATA, buf, 750 PATH_MAX - auio.uio_resid); 751 } 752 kfree(buf, M_JOURNAL); 753 } 754 done: 755 if (error) 756 jrecord_leaf(jrec, JLEAF_ERROR, &error, sizeof(error)); 757 jrecord_pop(jrec, save1); 758 } 759 760 /************************************************************************ 761 * JOURNAL VNOPS * 762 ************************************************************************ 763 * 764 * These are function shims replacing the normal filesystem ops. We become 765 * responsible for calling the underlying filesystem ops. We have the choice 766 * of executing the underlying op first and then generating the journal entry, 767 * or starting the journal entry, executing the underlying op, and then 768 * either completing or aborting it. 769 * 770 * The journal is supposed to be a high-level entity, which generally means 771 * identifying files by name rather then by inode. Supplying both allows 772 * the journal to be used both for inode-number-compatible 'mirrors' and 773 * for simple filesystem replication. 774 * 775 * Writes are particularly difficult to deal with because a single write may 776 * represent a hundred megabyte buffer or more, and both writes and truncations 777 * require the 'old' data to be written out as well as the new data if the 778 * log is reversable. Other issues: 779 * 780 * - How to deal with operations on unlinked files (no path available), 781 * but which may still be filesystem visible due to hard links. 782 * 783 * - How to deal with modifications made via a memory map. 784 * 785 * - Future cache coherency support will require cache coherency API calls 786 * both prior to and after the call to the underlying VFS. 787 * 788 * ALSO NOTE: We do not have to shim compatibility VOPs like MKDIR which have 789 * new VFS equivalents (NMKDIR). 790 */ 791 792 /* 793 * Journal vop_setattr { a_vp, a_vap, a_cred } 794 */ 795 static 796 int 797 journal_setattr(struct vop_setattr_args *ap) 798 { 799 struct jrecord_list jreclist; 800 struct jrecord jreccache; 801 struct jrecord *jrec; 802 struct mount *mp; 803 void *save; 804 int error; 805 806 mp = ap->a_head.a_ops->head.vv_mount; 807 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETATTR)) { 808 jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_VATTR, 0, 0); 809 } 810 error = vop_journal_operate_ap(&ap->a_head); 811 if (error == 0) { 812 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 813 jrecord_write_cred(jrec, curthread, ap->a_cred); 814 jrecord_write_vnode_ref(jrec, ap->a_vp); 815 save = jrecord_push(jrec, JTYPE_REDO); 816 jrecord_write_vattr(jrec, ap->a_vap); 817 jrecord_pop(jrec, save); 818 } 819 } 820 jreclist_done(mp, &jreclist, error); 821 return (error); 822 } 823 824 /* 825 * Journal vop_write { a_vp, a_uio, a_ioflag, a_cred } 826 */ 827 static 828 int 829 journal_write(struct vop_write_args *ap) 830 { 831 struct jrecord_list jreclist; 832 struct jrecord jreccache; 833 struct jrecord *jrec; 834 struct mount *mp; 835 struct uio uio_copy; 836 struct iovec uio_one_iovec; 837 void *save; 838 int error; 839 840 /* 841 * Special synchronizing writes for VM backing store do not supply any 842 * real data 843 */ 844 if (ap->a_uio->uio_segflg == UIO_NOCOPY) { 845 error = vop_journal_operate_ap(&ap->a_head); 846 return (error); 847 } 848 849 /* 850 * This is really nasty. UIO's don't retain sufficient information to 851 * be reusable once they've gone through the VOP chain. The iovecs get 852 * cleared, so we have to copy the UIO. 853 * 854 * XXX fix the UIO code to not destroy iov's during a scan so we can 855 * reuse the uio over and over again. 856 * 857 * XXX UNDO code needs to journal the old data prior to the write. 858 */ 859 uio_copy = *ap->a_uio; 860 if (uio_copy.uio_iovcnt == 1) { 861 uio_one_iovec = ap->a_uio->uio_iov[0]; 862 uio_copy.uio_iov = &uio_one_iovec; 863 } else { 864 uio_copy.uio_iov = kmalloc(uio_copy.uio_iovcnt * sizeof(struct iovec), 865 M_JOURNAL, M_WAITOK); 866 bcopy(ap->a_uio->uio_iov, uio_copy.uio_iov, 867 uio_copy.uio_iovcnt * sizeof(struct iovec)); 868 } 869 870 /* 871 * Write out undo data. Note that uio_offset is incorrect if 872 * IO_APPEND is set, but fortunately we have no undo file data to 873 * write out in that case. 874 */ 875 mp = ap->a_head.a_ops->head.vv_mount; 876 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_WRITE)) { 877 if (ap->a_ioflag & IO_APPEND) { 878 jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_SIZE|JRUNDO_MTIME, 0, 0); 879 } else { 880 jreclist_undo_file(&jreclist, ap->a_vp, 881 JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME, 882 uio_copy.uio_offset, uio_copy.uio_resid); 883 } 884 } 885 error = vop_journal_operate_ap(&ap->a_head); 886 887 /* 888 * XXX bad hack to figure out the offset for O_APPEND writes (note: 889 * uio field state after the VFS operation). 890 */ 891 uio_copy.uio_offset = ap->a_uio->uio_offset - 892 (uio_copy.uio_resid - ap->a_uio->uio_resid); 893 894 /* 895 * Output the write data to the journal. 896 */ 897 if (error == 0) { 898 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 899 jrecord_write_cred(jrec, NULL, ap->a_cred); 900 jrecord_write_vnode_ref(jrec, ap->a_vp); 901 save = jrecord_push(jrec, JTYPE_REDO); 902 jrecord_write_uio(jrec, JLEAF_FILEDATA, &uio_copy); 903 jrecord_pop(jrec, save); 904 } 905 } 906 jreclist_done(mp, &jreclist, error); 907 908 if (uio_copy.uio_iov != &uio_one_iovec) 909 kfree(uio_copy.uio_iov, M_JOURNAL); 910 return (error); 911 } 912 913 /* 914 * Journal vop_fsync { a_vp, a_waitfor } 915 */ 916 static 917 int 918 journal_fsync(struct vop_fsync_args *ap) 919 { 920 #if 0 921 struct mount *mp; 922 struct journal *jo; 923 #endif 924 int error; 925 926 error = vop_journal_operate_ap(&ap->a_head); 927 #if 0 928 mp = ap->a_head.a_ops->head.vv_mount; 929 if (error == 0) { 930 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) { 931 /* XXX synchronize pending journal records */ 932 } 933 } 934 #endif 935 return (error); 936 } 937 938 /* 939 * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset } 940 * 941 * note: a_count is in bytes. 942 */ 943 static 944 int 945 journal_putpages(struct vop_putpages_args *ap) 946 { 947 struct jrecord_list jreclist; 948 struct jrecord jreccache; 949 struct jrecord *jrec; 950 struct mount *mp; 951 void *save; 952 int error; 953 954 mp = ap->a_head.a_ops->head.vv_mount; 955 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_PUTPAGES) && 956 ap->a_count > 0 957 ) { 958 jreclist_undo_file(&jreclist, ap->a_vp, 959 JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME, 960 ap->a_offset, btoc(ap->a_count)); 961 } 962 error = vop_journal_operate_ap(&ap->a_head); 963 if (error == 0 && ap->a_count > 0) { 964 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 965 jrecord_write_vnode_ref(jrec, ap->a_vp); 966 save = jrecord_push(jrec, JTYPE_REDO); 967 jrecord_write_pagelist(jrec, JLEAF_FILEDATA, ap->a_m, ap->a_rtvals, 968 btoc(ap->a_count), ap->a_offset); 969 jrecord_pop(jrec, save); 970 } 971 } 972 jreclist_done(mp, &jreclist, error); 973 return (error); 974 } 975 976 /* 977 * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred } 978 */ 979 static 980 int 981 journal_setacl(struct vop_setacl_args *ap) 982 { 983 struct jrecord_list jreclist; 984 struct jrecord jreccache; 985 struct jrecord *jrec; 986 struct mount *mp; 987 int error; 988 989 mp = ap->a_head.a_ops->head.vv_mount; 990 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETACL); 991 error = vop_journal_operate_ap(&ap->a_head); 992 if (error == 0) { 993 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 994 #if 0 995 if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE)) 996 jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0); 997 #endif 998 jrecord_write_cred(jrec, curthread, ap->a_cred); 999 jrecord_write_vnode_ref(jrec, ap->a_vp); 1000 #if 0 1001 save = jrecord_push(jrec, JTYPE_REDO); 1002 /* XXX type, aclp */ 1003 jrecord_pop(jrec, save); 1004 #endif 1005 } 1006 } 1007 jreclist_done(mp, &jreclist, error); 1008 return (error); 1009 } 1010 1011 /* 1012 * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred } 1013 */ 1014 static 1015 int 1016 journal_setextattr(struct vop_setextattr_args *ap) 1017 { 1018 struct jrecord_list jreclist; 1019 struct jrecord jreccache; 1020 struct jrecord *jrec; 1021 struct mount *mp; 1022 void *save; 1023 int error; 1024 1025 mp = ap->a_head.a_ops->head.vv_mount; 1026 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETEXTATTR); 1027 error = vop_journal_operate_ap(&ap->a_head); 1028 if (error == 0) { 1029 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1030 #if 0 1031 if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE)) 1032 jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0); 1033 #endif 1034 jrecord_write_cred(jrec, curthread, ap->a_cred); 1035 jrecord_write_vnode_ref(jrec, ap->a_vp); 1036 jrecord_leaf(jrec, JLEAF_ATTRNAME, ap->a_attrname, 1037 strlen(ap->a_attrname)); 1038 save = jrecord_push(jrec, JTYPE_REDO); 1039 jrecord_write_uio(jrec, JLEAF_FILEDATA, ap->a_uio); 1040 jrecord_pop(jrec, save); 1041 } 1042 } 1043 jreclist_done(mp, &jreclist, error); 1044 return (error); 1045 } 1046 1047 /* 1048 * Journal vop_ncreate { a_nch, a_vpp, a_cred, a_vap } 1049 */ 1050 static 1051 int 1052 journal_ncreate(struct vop_ncreate_args *ap) 1053 { 1054 struct jrecord_list jreclist; 1055 struct jrecord jreccache; 1056 struct jrecord *jrec; 1057 struct mount *mp; 1058 void *save; 1059 int error; 1060 1061 mp = ap->a_head.a_ops->head.vv_mount; 1062 jreclist_init(mp, &jreclist, &jreccache, JTYPE_CREATE); 1063 error = vop_journal_operate_ap(&ap->a_head); 1064 if (error == 0) { 1065 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1066 jrecord_write_cred(jrec, NULL, ap->a_cred); 1067 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1068 if (*ap->a_vpp) 1069 jrecord_write_vnode_ref(jrec, *ap->a_vpp); 1070 save = jrecord_push(jrec, JTYPE_REDO); 1071 jrecord_write_vattr(jrec, ap->a_vap); 1072 jrecord_pop(jrec, save); 1073 } 1074 } 1075 jreclist_done(mp, &jreclist, error); 1076 return (error); 1077 } 1078 1079 /* 1080 * Journal vop_nmknod { a_nch, a_vpp, a_cred, a_vap } 1081 */ 1082 static 1083 int 1084 journal_nmknod(struct vop_nmknod_args *ap) 1085 { 1086 struct jrecord_list jreclist; 1087 struct jrecord jreccache; 1088 struct jrecord *jrec; 1089 struct mount *mp; 1090 void *save; 1091 int error; 1092 1093 mp = ap->a_head.a_ops->head.vv_mount; 1094 jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKNOD); 1095 error = vop_journal_operate_ap(&ap->a_head); 1096 if (error == 0) { 1097 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1098 jrecord_write_cred(jrec, NULL, ap->a_cred); 1099 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1100 save = jrecord_push(jrec, JTYPE_REDO); 1101 jrecord_write_vattr(jrec, ap->a_vap); 1102 jrecord_pop(jrec, save); 1103 if (*ap->a_vpp) 1104 jrecord_write_vnode_ref(jrec, *ap->a_vpp); 1105 } 1106 } 1107 jreclist_done(mp, &jreclist, error); 1108 return (error); 1109 } 1110 1111 /* 1112 * Journal vop_nlink { a_nch, a_vp, a_cred } 1113 */ 1114 static 1115 int 1116 journal_nlink(struct vop_nlink_args *ap) 1117 { 1118 struct jrecord_list jreclist; 1119 struct jrecord jreccache; 1120 struct jrecord *jrec; 1121 struct mount *mp; 1122 void *save; 1123 int error; 1124 1125 mp = ap->a_head.a_ops->head.vv_mount; 1126 jreclist_init(mp, &jreclist, &jreccache, JTYPE_LINK); 1127 error = vop_journal_operate_ap(&ap->a_head); 1128 if (error == 0) { 1129 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1130 jrecord_write_cred(jrec, NULL, ap->a_cred); 1131 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1132 /* XXX PATH to VP and inode number */ 1133 /* XXX this call may not record the correct path when 1134 * multiple paths are available */ 1135 save = jrecord_push(jrec, JTYPE_REDO); 1136 jrecord_write_vnode_link(jrec, ap->a_vp, ap->a_nch->ncp); 1137 jrecord_pop(jrec, save); 1138 } 1139 } 1140 jreclist_done(mp, &jreclist, error); 1141 return (error); 1142 } 1143 1144 /* 1145 * Journal vop_symlink { a_nch, a_vpp, a_cred, a_vap, a_target } 1146 */ 1147 static 1148 int 1149 journal_nsymlink(struct vop_nsymlink_args *ap) 1150 { 1151 struct jrecord_list jreclist; 1152 struct jrecord jreccache; 1153 struct jrecord *jrec; 1154 struct mount *mp; 1155 void *save; 1156 int error; 1157 1158 mp = ap->a_head.a_ops->head.vv_mount; 1159 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SYMLINK); 1160 error = vop_journal_operate_ap(&ap->a_head); 1161 if (error == 0) { 1162 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1163 jrecord_write_cred(jrec, NULL, ap->a_cred); 1164 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1165 save = jrecord_push(jrec, JTYPE_REDO); 1166 jrecord_leaf(jrec, JLEAF_SYMLINKDATA, 1167 ap->a_target, strlen(ap->a_target)); 1168 jrecord_pop(jrec, save); 1169 if (*ap->a_vpp) 1170 jrecord_write_vnode_ref(jrec, *ap->a_vpp); 1171 } 1172 } 1173 jreclist_done(mp, &jreclist, error); 1174 return (error); 1175 } 1176 1177 /* 1178 * Journal vop_nwhiteout { a_nch, a_cred, a_flags } 1179 */ 1180 static 1181 int 1182 journal_nwhiteout(struct vop_nwhiteout_args *ap) 1183 { 1184 struct jrecord_list jreclist; 1185 struct jrecord jreccache; 1186 struct jrecord *jrec; 1187 struct mount *mp; 1188 int error; 1189 1190 mp = ap->a_head.a_ops->head.vv_mount; 1191 jreclist_init(mp, &jreclist, &jreccache, JTYPE_WHITEOUT); 1192 error = vop_journal_operate_ap(&ap->a_head); 1193 if (error == 0) { 1194 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1195 jrecord_write_cred(jrec, NULL, ap->a_cred); 1196 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1197 } 1198 } 1199 jreclist_done(mp, &jreclist, error); 1200 return (error); 1201 } 1202 1203 /* 1204 * Journal vop_nremove { a_nch, a_cred } 1205 */ 1206 static 1207 int 1208 journal_nremove(struct vop_nremove_args *ap) 1209 { 1210 struct jrecord_list jreclist; 1211 struct jrecord jreccache; 1212 struct jrecord *jrec; 1213 struct mount *mp; 1214 int error; 1215 1216 mp = ap->a_head.a_ops->head.vv_mount; 1217 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_REMOVE) && 1218 ap->a_nch->ncp->nc_vp 1219 ) { 1220 jreclist_undo_file(&jreclist, ap->a_nch->ncp->nc_vp, 1221 JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1); 1222 } 1223 error = vop_journal_operate_ap(&ap->a_head); 1224 if (error == 0) { 1225 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1226 jrecord_write_cred(jrec, NULL, ap->a_cred); 1227 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1228 } 1229 } 1230 jreclist_done(mp, &jreclist, error); 1231 return (error); 1232 } 1233 1234 /* 1235 * Journal vop_nmkdir { a_nch, a_vpp, a_cred, a_vap } 1236 */ 1237 static 1238 int 1239 journal_nmkdir(struct vop_nmkdir_args *ap) 1240 { 1241 struct jrecord_list jreclist; 1242 struct jrecord jreccache; 1243 struct jrecord *jrec; 1244 struct mount *mp; 1245 int error; 1246 1247 mp = ap->a_head.a_ops->head.vv_mount; 1248 jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKDIR); 1249 error = vop_journal_operate_ap(&ap->a_head); 1250 if (error == 0) { 1251 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1252 #if 0 1253 if (jo->flags & MC_JOURNAL_WANT_AUDIT) { 1254 jrecord_write_audit(jrec); 1255 } 1256 #endif 1257 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1258 jrecord_write_cred(jrec, NULL, ap->a_cred); 1259 jrecord_write_vattr(jrec, ap->a_vap); 1260 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1261 if (*ap->a_vpp) 1262 jrecord_write_vnode_ref(jrec, *ap->a_vpp); 1263 } 1264 } 1265 jreclist_done(mp, &jreclist, error); 1266 return (error); 1267 } 1268 1269 /* 1270 * Journal vop_nrmdir { a_nch, a_cred } 1271 */ 1272 static 1273 int 1274 journal_nrmdir(struct vop_nrmdir_args *ap) 1275 { 1276 struct jrecord_list jreclist; 1277 struct jrecord jreccache; 1278 struct jrecord *jrec; 1279 struct mount *mp; 1280 int error; 1281 1282 mp = ap->a_head.a_ops->head.vv_mount; 1283 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RMDIR)) { 1284 jreclist_undo_file(&jreclist, ap->a_nch->ncp->nc_vp, 1285 JRUNDO_VATTR|JRUNDO_GETVP, 0, 0); 1286 } 1287 error = vop_journal_operate_ap(&ap->a_head); 1288 if (error == 0) { 1289 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1290 jrecord_write_cred(jrec, NULL, ap->a_cred); 1291 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp); 1292 } 1293 } 1294 jreclist_done(mp, &jreclist, error); 1295 return (error); 1296 } 1297 1298 /* 1299 * Journal vop_nrename { a_fnch, a_tnch, a_cred } 1300 */ 1301 static 1302 int 1303 journal_nrename(struct vop_nrename_args *ap) 1304 { 1305 struct jrecord_list jreclist; 1306 struct jrecord jreccache; 1307 struct jrecord *jrec; 1308 struct mount *mp; 1309 int error; 1310 1311 mp = ap->a_head.a_ops->head.vv_mount; 1312 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RENAME) && 1313 ap->a_tnch->ncp->nc_vp 1314 ) { 1315 jreclist_undo_file(&jreclist, ap->a_tnch->ncp->nc_vp, 1316 JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1); 1317 } 1318 error = vop_journal_operate_ap(&ap->a_head); 1319 if (error == 0) { 1320 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) { 1321 jrecord_write_cred(jrec, NULL, ap->a_cred); 1322 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_fnch->ncp); 1323 jrecord_write_path(jrec, JLEAF_PATH2, ap->a_tnch->ncp); 1324 } 1325 } 1326 jreclist_done(mp, &jreclist, error); 1327 return (error); 1328 } 1329 1330