1 /* $NetBSD: fss.c,v 1.87 2014/02/28 10:16:51 skrll Exp $ */ 2 3 /*- 4 * Copyright (c) 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Juergen Hannken-Illjes. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * File system snapshot disk driver. 34 * 35 * Block/character interface to the snapshot of a mounted file system. 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.87 2014/02/28 10:16:51 skrll Exp $"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/namei.h> 44 #include <sys/proc.h> 45 #include <sys/errno.h> 46 #include <sys/malloc.h> 47 #include <sys/buf.h> 48 #include <sys/ioctl.h> 49 #include <sys/disklabel.h> 50 #include <sys/device.h> 51 #include <sys/disk.h> 52 #include <sys/stat.h> 53 #include <sys/mount.h> 54 #include <sys/vnode.h> 55 #include <sys/file.h> 56 #include <sys/uio.h> 57 #include <sys/conf.h> 58 #include <sys/kthread.h> 59 #include <sys/fstrans.h> 60 #include <sys/vfs_syscalls.h> /* For do_sys_unlink(). */ 61 62 #include <miscfs/specfs/specdev.h> 63 64 #include <dev/fssvar.h> 65 66 #include <uvm/uvm.h> 67 68 void fssattach(int); 69 70 dev_type_open(fss_open); 71 dev_type_close(fss_close); 72 dev_type_read(fss_read); 73 dev_type_write(fss_write); 74 dev_type_ioctl(fss_ioctl); 75 dev_type_strategy(fss_strategy); 76 dev_type_dump(fss_dump); 77 dev_type_size(fss_size); 78 79 static void fss_unmount_hook(struct mount *); 80 static int fss_copy_on_write(void *, struct buf *, bool); 81 static inline void fss_error(struct fss_softc *, const char *); 82 static int fss_create_files(struct fss_softc *, struct fss_set *, 83 off_t *, struct lwp *); 84 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 85 struct lwp *); 86 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 87 static int fss_softc_alloc(struct fss_softc *); 88 static void fss_softc_free(struct fss_softc *); 89 static int fss_read_cluster(struct fss_softc *, u_int32_t); 90 static void fss_bs_thread(void *); 91 static int fss_bs_io(struct fss_softc *, fss_io_type, 92 u_int32_t, off_t, int, void *); 93 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 94 95 static kmutex_t fss_device_lock; /* Protect all units. */ 96 static int fss_num_attached = 0; /* Number of attached devices. */ 97 static struct vfs_hooks fss_vfs_hooks = { 98 .vh_unmount = fss_unmount_hook 99 }; 100 101 const struct bdevsw fss_bdevsw = { 102 fss_open, fss_close, fss_strategy, fss_ioctl, 103 fss_dump, fss_size, D_DISK | D_MPSAFE 104 }; 105 106 const struct cdevsw fss_cdevsw = { 107 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 108 nostop, notty, nopoll, nommap, nokqfilter, D_DISK | D_MPSAFE 109 }; 110 111 static int fss_match(device_t, cfdata_t, void *); 112 static void fss_attach(device_t, device_t, void *); 113 static int fss_detach(device_t, int); 114 115 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 116 fss_match, fss_attach, fss_detach, NULL); 117 extern struct cfdriver fss_cd; 118 119 void 120 fssattach(int num) 121 { 122 123 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 124 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 125 aprint_error("%s: unable to register\n", fss_cd.cd_name); 126 } 127 128 static int 129 fss_match(device_t self, cfdata_t cfdata, void *aux) 130 { 131 return 1; 132 } 133 134 static void 135 fss_attach(device_t parent, device_t self, void *aux) 136 { 137 struct fss_softc *sc = device_private(self); 138 139 sc->sc_dev = self; 140 sc->sc_bdev = NODEV; 141 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 142 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 143 cv_init(&sc->sc_work_cv, "fssbs"); 144 cv_init(&sc->sc_cache_cv, "cowwait"); 145 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 146 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 147 sc->sc_dkdev->dk_info = NULL; 148 disk_init(sc->sc_dkdev, device_xname(self), NULL); 149 if (!pmf_device_register(self, NULL, NULL)) 150 aprint_error_dev(self, "couldn't establish power handler\n"); 151 152 if (fss_num_attached++ == 0) 153 vfs_hooks_attach(&fss_vfs_hooks); 154 } 155 156 static int 157 fss_detach(device_t self, int flags) 158 { 159 struct fss_softc *sc = device_private(self); 160 161 if (sc->sc_flags & FSS_ACTIVE) 162 return EBUSY; 163 164 if (--fss_num_attached == 0) 165 vfs_hooks_detach(&fss_vfs_hooks); 166 167 pmf_device_deregister(self); 168 mutex_destroy(&sc->sc_slock); 169 mutex_destroy(&sc->sc_lock); 170 cv_destroy(&sc->sc_work_cv); 171 cv_destroy(&sc->sc_cache_cv); 172 bufq_drain(sc->sc_bufq); 173 bufq_free(sc->sc_bufq); 174 disk_destroy(sc->sc_dkdev); 175 free(sc->sc_dkdev, M_DEVBUF); 176 177 return 0; 178 } 179 180 int 181 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 182 { 183 int mflag; 184 cfdata_t cf; 185 struct fss_softc *sc; 186 187 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 188 189 mutex_enter(&fss_device_lock); 190 191 sc = device_lookup_private(&fss_cd, minor(dev)); 192 if (sc == NULL) { 193 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 194 cf->cf_name = fss_cd.cd_name; 195 cf->cf_atname = fss_cd.cd_name; 196 cf->cf_unit = minor(dev); 197 cf->cf_fstate = FSTATE_STAR; 198 sc = device_private(config_attach_pseudo(cf)); 199 if (sc == NULL) { 200 mutex_exit(&fss_device_lock); 201 return ENOMEM; 202 } 203 } 204 205 mutex_enter(&sc->sc_slock); 206 207 sc->sc_flags |= mflag; 208 209 mutex_exit(&sc->sc_slock); 210 mutex_exit(&fss_device_lock); 211 212 return 0; 213 } 214 215 int 216 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 217 { 218 int mflag, error; 219 cfdata_t cf; 220 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 221 222 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 223 error = 0; 224 225 mutex_enter(&fss_device_lock); 226 restart: 227 mutex_enter(&sc->sc_slock); 228 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 229 sc->sc_flags &= ~mflag; 230 mutex_exit(&sc->sc_slock); 231 mutex_exit(&fss_device_lock); 232 return 0; 233 } 234 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 235 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 236 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 237 mutex_exit(&sc->sc_slock); 238 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 239 goto restart; 240 } 241 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 242 mutex_exit(&sc->sc_slock); 243 mutex_exit(&fss_device_lock); 244 return error; 245 } 246 247 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 248 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 249 mutex_exit(&sc->sc_slock); 250 cf = device_cfdata(sc->sc_dev); 251 error = config_detach(sc->sc_dev, DETACH_QUIET); 252 if (! error) 253 free(cf, M_DEVBUF); 254 mutex_exit(&fss_device_lock); 255 256 return error; 257 } 258 259 void 260 fss_strategy(struct buf *bp) 261 { 262 const bool write = ((bp->b_flags & B_READ) != B_READ); 263 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 264 265 mutex_enter(&sc->sc_slock); 266 267 if (write || !FSS_ISVALID(sc)) { 268 269 mutex_exit(&sc->sc_slock); 270 271 bp->b_error = (write ? EROFS : ENXIO); 272 bp->b_resid = bp->b_bcount; 273 biodone(bp); 274 return; 275 } 276 277 bp->b_rawblkno = bp->b_blkno; 278 bufq_put(sc->sc_bufq, bp); 279 cv_signal(&sc->sc_work_cv); 280 281 mutex_exit(&sc->sc_slock); 282 } 283 284 int 285 fss_read(dev_t dev, struct uio *uio, int flags) 286 { 287 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 288 } 289 290 int 291 fss_write(dev_t dev, struct uio *uio, int flags) 292 { 293 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 294 } 295 296 int 297 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 298 { 299 int error; 300 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 301 struct fss_set _fss; 302 struct fss_set *fss = (struct fss_set *)data; 303 struct fss_set50 *fss50 = (struct fss_set50 *)data; 304 struct fss_get *fsg = (struct fss_get *)data; 305 #ifndef _LP64 306 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 307 #endif 308 309 switch (cmd) { 310 case FSSIOCSET50: 311 fss = &_fss; 312 fss->fss_mount = fss50->fss_mount; 313 fss->fss_bstore = fss50->fss_bstore; 314 fss->fss_csize = fss50->fss_csize; 315 fss->fss_flags = 0; 316 /* Fall through */ 317 case FSSIOCSET: 318 mutex_enter(&sc->sc_lock); 319 if ((flag & FWRITE) == 0) 320 error = EPERM; 321 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 322 error = EBUSY; 323 else 324 error = fss_create_snapshot(sc, fss, l); 325 if (error == 0) 326 sc->sc_uflags = fss->fss_flags; 327 mutex_exit(&sc->sc_lock); 328 break; 329 330 case FSSIOCCLR: 331 mutex_enter(&sc->sc_lock); 332 if ((flag & FWRITE) == 0) 333 error = EPERM; 334 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 335 error = ENXIO; 336 else 337 error = fss_delete_snapshot(sc, l); 338 mutex_exit(&sc->sc_lock); 339 break; 340 341 #ifndef _LP64 342 case FSSIOCGET50: 343 mutex_enter(&sc->sc_lock); 344 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 345 case FSS_ACTIVE: 346 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 347 fsg50->fsg_csize = FSS_CLSIZE(sc); 348 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 349 fsg50->fsg_mount_size = sc->sc_clcount; 350 fsg50->fsg_bs_size = sc->sc_clnext; 351 error = 0; 352 break; 353 case FSS_PERSISTENT | FSS_ACTIVE: 354 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 355 fsg50->fsg_csize = 0; 356 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 357 fsg50->fsg_mount_size = 0; 358 fsg50->fsg_bs_size = 0; 359 error = 0; 360 break; 361 default: 362 error = ENXIO; 363 break; 364 } 365 mutex_exit(&sc->sc_lock); 366 break; 367 #endif /* _LP64 */ 368 369 case FSSIOCGET: 370 mutex_enter(&sc->sc_lock); 371 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 372 case FSS_ACTIVE: 373 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 374 fsg->fsg_csize = FSS_CLSIZE(sc); 375 fsg->fsg_time = sc->sc_time; 376 fsg->fsg_mount_size = sc->sc_clcount; 377 fsg->fsg_bs_size = sc->sc_clnext; 378 error = 0; 379 break; 380 case FSS_PERSISTENT | FSS_ACTIVE: 381 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 382 fsg->fsg_csize = 0; 383 fsg->fsg_time = sc->sc_time; 384 fsg->fsg_mount_size = 0; 385 fsg->fsg_bs_size = 0; 386 error = 0; 387 break; 388 default: 389 error = ENXIO; 390 break; 391 } 392 mutex_exit(&sc->sc_lock); 393 break; 394 395 case FSSIOFSET: 396 mutex_enter(&sc->sc_slock); 397 sc->sc_uflags = *(int *)data; 398 mutex_exit(&sc->sc_slock); 399 error = 0; 400 break; 401 402 case FSSIOFGET: 403 mutex_enter(&sc->sc_slock); 404 *(int *)data = sc->sc_uflags; 405 mutex_exit(&sc->sc_slock); 406 error = 0; 407 break; 408 409 default: 410 error = EINVAL; 411 break; 412 } 413 414 return error; 415 } 416 417 int 418 fss_size(dev_t dev) 419 { 420 return -1; 421 } 422 423 int 424 fss_dump(dev_t dev, daddr_t blkno, void *va, 425 size_t size) 426 { 427 return EROFS; 428 } 429 430 /* 431 * An error occurred reading or writing the snapshot or backing store. 432 * If it is the first error log to console. 433 * The caller holds the mutex. 434 */ 435 static inline void 436 fss_error(struct fss_softc *sc, const char *msg) 437 { 438 439 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 440 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 441 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 442 sc->sc_flags |= FSS_ERROR; 443 } 444 445 /* 446 * Allocate the variable sized parts of the softc and 447 * fork the kernel thread. 448 * 449 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 450 * must be initialized. 451 */ 452 static int 453 fss_softc_alloc(struct fss_softc *sc) 454 { 455 int i, error; 456 457 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 458 sc->sc_copied = 459 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 460 if (sc->sc_copied == NULL) 461 return(ENOMEM); 462 463 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 464 sizeof(struct fss_cache), KM_SLEEP); 465 if (sc->sc_cache == NULL) 466 return(ENOMEM); 467 468 for (i = 0; i < sc->sc_cache_size; i++) { 469 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 470 sc->sc_cache[i].fc_data = 471 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 472 if (sc->sc_cache[i].fc_data == NULL) 473 return(ENOMEM); 474 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 475 } 476 477 sc->sc_indir_valid = 478 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 479 if (sc->sc_indir_valid == NULL) 480 return(ENOMEM); 481 482 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 483 if (sc->sc_indir_data == NULL) 484 return(ENOMEM); 485 } else { 486 sc->sc_copied = NULL; 487 sc->sc_cache = NULL; 488 sc->sc_indir_valid = NULL; 489 sc->sc_indir_data = NULL; 490 } 491 492 sc->sc_flags |= FSS_BS_THREAD; 493 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 494 fss_bs_thread, sc, &sc->sc_bs_lwp, 495 "%s", device_xname(sc->sc_dev))) != 0) { 496 sc->sc_flags &= ~FSS_BS_THREAD; 497 return error; 498 } 499 500 disk_attach(sc->sc_dkdev); 501 502 return 0; 503 } 504 505 /* 506 * Free the variable sized parts of the softc. 507 */ 508 static void 509 fss_softc_free(struct fss_softc *sc) 510 { 511 int i; 512 513 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 514 mutex_enter(&sc->sc_slock); 515 sc->sc_flags &= ~FSS_BS_THREAD; 516 cv_signal(&sc->sc_work_cv); 517 mutex_exit(&sc->sc_slock); 518 kthread_join(sc->sc_bs_lwp); 519 520 disk_detach(sc->sc_dkdev); 521 } 522 523 if (sc->sc_copied != NULL) 524 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 525 sc->sc_copied = NULL; 526 527 if (sc->sc_cache != NULL) { 528 for (i = 0; i < sc->sc_cache_size; i++) 529 if (sc->sc_cache[i].fc_data != NULL) { 530 cv_destroy(&sc->sc_cache[i].fc_state_cv); 531 kmem_free(sc->sc_cache[i].fc_data, 532 FSS_CLSIZE(sc)); 533 } 534 kmem_free(sc->sc_cache, 535 sc->sc_cache_size*sizeof(struct fss_cache)); 536 } 537 sc->sc_cache = NULL; 538 539 if (sc->sc_indir_valid != NULL) 540 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 541 sc->sc_indir_valid = NULL; 542 543 if (sc->sc_indir_data != NULL) 544 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 545 sc->sc_indir_data = NULL; 546 } 547 548 /* 549 * Set all active snapshots on this file system into ERROR state. 550 */ 551 static void 552 fss_unmount_hook(struct mount *mp) 553 { 554 int i; 555 struct fss_softc *sc; 556 557 mutex_enter(&fss_device_lock); 558 for (i = 0; i < fss_cd.cd_ndevs; i++) { 559 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 560 continue; 561 mutex_enter(&sc->sc_slock); 562 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 563 sc->sc_mount == mp) 564 fss_error(sc, "forced unmount"); 565 mutex_exit(&sc->sc_slock); 566 } 567 mutex_exit(&fss_device_lock); 568 } 569 570 /* 571 * A buffer is written to the snapshotted block device. Copy to 572 * backing store if needed. 573 */ 574 static int 575 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 576 { 577 int error; 578 u_int32_t cl, ch, c; 579 struct fss_softc *sc = v; 580 581 mutex_enter(&sc->sc_slock); 582 if (!FSS_ISVALID(sc)) { 583 mutex_exit(&sc->sc_slock); 584 return 0; 585 } 586 587 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 588 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 589 error = 0; 590 if (curlwp == uvm.pagedaemon_lwp) { 591 for (c = cl; c <= ch; c++) 592 if (isclr(sc->sc_copied, c)) { 593 error = ENOMEM; 594 break; 595 } 596 } 597 mutex_exit(&sc->sc_slock); 598 599 if (error == 0) 600 for (c = cl; c <= ch; c++) { 601 error = fss_read_cluster(sc, c); 602 if (error) 603 break; 604 } 605 606 return error; 607 } 608 609 /* 610 * Lookup and open needed files. 611 * 612 * For file system internal snapshot initializes sc_mntname, sc_mount, 613 * sc_bs_vp and sc_time. 614 * 615 * Otherwise returns dev and size of the underlying block device. 616 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 617 */ 618 static int 619 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 620 off_t *bsize, struct lwp *l) 621 { 622 int error, bits, fsbsize; 623 uint64_t numsec; 624 unsigned int secsize; 625 struct timespec ts; 626 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 627 struct pathbuf *pb2; 628 struct nameidata nd2; 629 struct vnode *vp; 630 631 /* 632 * Get the mounted file system. 633 */ 634 635 error = namei_simple_user(fss->fss_mount, 636 NSM_FOLLOW_NOEMULROOT, &vp); 637 if (error != 0) 638 return error; 639 640 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 641 vrele(vp); 642 return EINVAL; 643 } 644 645 sc->sc_mount = vp->v_mount; 646 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 647 648 vrele(vp); 649 650 /* 651 * Check for file system internal snapshot. 652 */ 653 654 error = namei_simple_user(fss->fss_bstore, 655 NSM_FOLLOW_NOEMULROOT, &vp); 656 if (error != 0) 657 return error; 658 659 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 660 sc->sc_flags |= FSS_PERSISTENT; 661 sc->sc_bs_vp = vp; 662 663 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 664 bits = sizeof(sc->sc_bs_bshift)*NBBY; 665 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 666 sc->sc_bs_bshift++) 667 if (FSS_FSBSIZE(sc) == fsbsize) 668 break; 669 if (sc->sc_bs_bshift >= bits) 670 return EINVAL; 671 672 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 673 sc->sc_clshift = 0; 674 675 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 676 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 677 if (error) 678 return error; 679 } 680 error = vn_lock(vp, LK_EXCLUSIVE); 681 if (error != 0) 682 return error; 683 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 684 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 685 686 VOP_UNLOCK(sc->sc_bs_vp); 687 688 return error; 689 } 690 vrele(vp); 691 692 /* 693 * Get the block device it is mounted on and its size. 694 */ 695 696 error = spec_node_lookup_by_mount(sc->sc_mount, &vp); 697 if (error) 698 return error; 699 sc->sc_bdev = vp->v_rdev; 700 701 error = getdisksize(vp, &numsec, &secsize); 702 vrele(vp); 703 if (error) 704 return error; 705 706 *bsize = (off_t)numsec*secsize; 707 708 /* 709 * Get the backing store 710 */ 711 712 error = pathbuf_copyin(fss->fss_bstore, &pb2); 713 if (error) { 714 return error; 715 } 716 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 717 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 718 pathbuf_destroy(pb2); 719 return error; 720 } 721 VOP_UNLOCK(nd2.ni_vp); 722 723 sc->sc_bs_vp = nd2.ni_vp; 724 725 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 726 pathbuf_destroy(pb2); 727 return EINVAL; 728 } 729 pathbuf_destroy(pb2); 730 731 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 732 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 733 if (error) 734 return error; 735 } 736 if (sc->sc_bs_vp->v_type == VREG) { 737 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 738 if (fsbsize & (fsbsize-1)) /* No power of two */ 739 return EINVAL; 740 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 741 sc->sc_bs_bshift++) 742 if (FSS_FSBSIZE(sc) == fsbsize) 743 break; 744 if (sc->sc_bs_bshift >= 32) 745 return EINVAL; 746 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 747 } else { 748 sc->sc_bs_bshift = DEV_BSHIFT; 749 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 750 } 751 752 return 0; 753 } 754 755 /* 756 * Create a snapshot. 757 */ 758 static int 759 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 760 { 761 int len, error; 762 u_int32_t csize; 763 off_t bsize; 764 765 bsize = 0; /* XXX gcc */ 766 767 /* 768 * Open needed files. 769 */ 770 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 771 goto bad; 772 773 if (sc->sc_flags & FSS_PERSISTENT) { 774 fss_softc_alloc(sc); 775 sc->sc_flags |= FSS_ACTIVE; 776 return 0; 777 } 778 779 /* 780 * Set cluster size. Must be a power of two and 781 * a multiple of backing store block size. 782 */ 783 if (fss->fss_csize <= 0) 784 csize = MAXPHYS; 785 else 786 csize = fss->fss_csize; 787 if (bsize/csize > FSS_CLUSTER_MAX) 788 csize = bsize/FSS_CLUSTER_MAX+1; 789 790 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 791 sc->sc_clshift++) 792 if (FSS_CLSIZE(sc) >= csize) 793 break; 794 if (sc->sc_clshift >= 32) { 795 error = EINVAL; 796 goto bad; 797 } 798 sc->sc_clmask = FSS_CLSIZE(sc)-1; 799 800 /* 801 * Set number of cache slots. 802 */ 803 if (FSS_CLSIZE(sc) <= 8192) 804 sc->sc_cache_size = 32; 805 else if (FSS_CLSIZE(sc) <= 65536) 806 sc->sc_cache_size = 8; 807 else 808 sc->sc_cache_size = 4; 809 810 /* 811 * Set number of clusters and size of last cluster. 812 */ 813 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 814 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 815 816 /* 817 * Set size of indirect table. 818 */ 819 len = sc->sc_clcount*sizeof(u_int32_t); 820 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 821 sc->sc_clnext = sc->sc_indir_size; 822 sc->sc_indir_cur = 0; 823 824 if ((error = fss_softc_alloc(sc)) != 0) 825 goto bad; 826 827 /* 828 * Activate the snapshot. 829 */ 830 831 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 832 goto bad; 833 834 microtime(&sc->sc_time); 835 836 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 837 if (error == 0) 838 sc->sc_flags |= FSS_ACTIVE; 839 840 vfs_resume(sc->sc_mount); 841 842 if (error != 0) 843 goto bad; 844 845 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 846 aprint_debug_dev(sc->sc_dev, 847 "%u clusters of %u, %u cache slots, %u indir clusters\n", 848 sc->sc_clcount, FSS_CLSIZE(sc), 849 sc->sc_cache_size, sc->sc_indir_size); 850 851 return 0; 852 853 bad: 854 fss_softc_free(sc); 855 if (sc->sc_bs_vp != NULL) { 856 if (sc->sc_flags & FSS_PERSISTENT) 857 vrele(sc->sc_bs_vp); 858 else 859 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 860 } 861 sc->sc_bs_vp = NULL; 862 863 return error; 864 } 865 866 /* 867 * Delete a snapshot. 868 */ 869 static int 870 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 871 { 872 873 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 874 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 875 876 mutex_enter(&sc->sc_slock); 877 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 878 sc->sc_mount = NULL; 879 sc->sc_bdev = NODEV; 880 mutex_exit(&sc->sc_slock); 881 882 fss_softc_free(sc); 883 if (sc->sc_flags & FSS_PERSISTENT) 884 vrele(sc->sc_bs_vp); 885 else 886 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 887 sc->sc_bs_vp = NULL; 888 sc->sc_flags &= ~FSS_PERSISTENT; 889 890 return 0; 891 } 892 893 /* 894 * Read a cluster from the snapshotted block device to the cache. 895 */ 896 static int 897 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 898 { 899 int error, todo, offset, len; 900 daddr_t dblk; 901 struct buf *bp, *mbp; 902 struct fss_cache *scp, *scl; 903 904 /* 905 * Get a free cache slot. 906 */ 907 scl = sc->sc_cache+sc->sc_cache_size; 908 909 mutex_enter(&sc->sc_slock); 910 911 restart: 912 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 913 mutex_exit(&sc->sc_slock); 914 return 0; 915 } 916 917 for (scp = sc->sc_cache; scp < scl; scp++) 918 if (scp->fc_cluster == cl) { 919 if (scp->fc_type == FSS_CACHE_VALID) { 920 mutex_exit(&sc->sc_slock); 921 return 0; 922 } else if (scp->fc_type == FSS_CACHE_BUSY) { 923 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 924 goto restart; 925 } 926 } 927 928 for (scp = sc->sc_cache; scp < scl; scp++) 929 if (scp->fc_type == FSS_CACHE_FREE) { 930 scp->fc_type = FSS_CACHE_BUSY; 931 scp->fc_cluster = cl; 932 break; 933 } 934 if (scp >= scl) { 935 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 936 goto restart; 937 } 938 939 mutex_exit(&sc->sc_slock); 940 941 /* 942 * Start the read. 943 */ 944 dblk = btodb(FSS_CLTOB(sc, cl)); 945 if (cl == sc->sc_clcount-1) { 946 todo = sc->sc_clresid; 947 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 948 } else 949 todo = FSS_CLSIZE(sc); 950 offset = 0; 951 mbp = getiobuf(NULL, true); 952 mbp->b_bufsize = todo; 953 mbp->b_data = scp->fc_data; 954 mbp->b_resid = mbp->b_bcount = todo; 955 mbp->b_flags = B_READ; 956 mbp->b_cflags = BC_BUSY; 957 mbp->b_dev = sc->sc_bdev; 958 while (todo > 0) { 959 len = todo; 960 if (len > MAXPHYS) 961 len = MAXPHYS; 962 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 963 bp = mbp; 964 else { 965 bp = getiobuf(NULL, true); 966 nestiobuf_setup(mbp, bp, offset, len); 967 } 968 bp->b_lblkno = 0; 969 bp->b_blkno = dblk; 970 bdev_strategy(bp); 971 dblk += btodb(len); 972 offset += len; 973 todo -= len; 974 } 975 error = biowait(mbp); 976 putiobuf(mbp); 977 978 mutex_enter(&sc->sc_slock); 979 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 980 cv_broadcast(&scp->fc_state_cv); 981 if (error == 0) { 982 setbit(sc->sc_copied, scp->fc_cluster); 983 cv_signal(&sc->sc_work_cv); 984 } 985 mutex_exit(&sc->sc_slock); 986 987 return error; 988 } 989 990 /* 991 * Read/write clusters from/to backing store. 992 * For persistent snapshots must be called with cl == 0. off is the 993 * offset into the snapshot. 994 */ 995 static int 996 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 997 u_int32_t cl, off_t off, int len, void *data) 998 { 999 int error; 1000 1001 off += FSS_CLTOB(sc, cl); 1002 1003 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1004 1005 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1006 data, len, off, UIO_SYSSPACE, 1007 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1008 sc->sc_bs_lwp->l_cred, NULL, NULL); 1009 if (error == 0) { 1010 mutex_enter(sc->sc_bs_vp->v_interlock); 1011 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1012 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1013 } 1014 1015 VOP_UNLOCK(sc->sc_bs_vp); 1016 1017 return error; 1018 } 1019 1020 /* 1021 * Get a pointer to the indirect slot for this cluster. 1022 */ 1023 static u_int32_t * 1024 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1025 { 1026 u_int32_t icl; 1027 int ioff; 1028 1029 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1030 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1031 1032 if (sc->sc_indir_cur == icl) 1033 return &sc->sc_indir_data[ioff]; 1034 1035 if (sc->sc_indir_dirty) { 1036 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1037 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1038 return NULL; 1039 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1040 } 1041 1042 sc->sc_indir_dirty = 0; 1043 sc->sc_indir_cur = icl; 1044 1045 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1046 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1047 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1048 return NULL; 1049 } else 1050 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1051 1052 return &sc->sc_indir_data[ioff]; 1053 } 1054 1055 /* 1056 * The kernel thread (one for every active snapshot). 1057 * 1058 * After wakeup it cleans the cache and runs the I/O requests. 1059 */ 1060 static void 1061 fss_bs_thread(void *arg) 1062 { 1063 bool thread_idle, is_valid; 1064 int error, i, todo, len, crotor, is_read; 1065 long off; 1066 char *addr; 1067 u_int32_t c, cl, ch, *indirp; 1068 struct buf *bp, *nbp; 1069 struct fss_softc *sc; 1070 struct fss_cache *scp, *scl; 1071 1072 sc = arg; 1073 scl = sc->sc_cache+sc->sc_cache_size; 1074 crotor = 0; 1075 thread_idle = false; 1076 1077 mutex_enter(&sc->sc_slock); 1078 1079 for (;;) { 1080 if (thread_idle) 1081 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1082 thread_idle = true; 1083 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1084 mutex_exit(&sc->sc_slock); 1085 kthread_exit(0); 1086 } 1087 1088 /* 1089 * Process I/O requests (persistent) 1090 */ 1091 1092 if (sc->sc_flags & FSS_PERSISTENT) { 1093 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1094 continue; 1095 is_valid = FSS_ISVALID(sc); 1096 is_read = (bp->b_flags & B_READ); 1097 thread_idle = false; 1098 mutex_exit(&sc->sc_slock); 1099 1100 if (is_valid) { 1101 disk_busy(sc->sc_dkdev); 1102 error = fss_bs_io(sc, FSS_READ, 0, 1103 dbtob(bp->b_blkno), bp->b_bcount, 1104 bp->b_data); 1105 disk_unbusy(sc->sc_dkdev, 1106 (error ? 0 : bp->b_bcount), is_read); 1107 } else 1108 error = ENXIO; 1109 1110 bp->b_error = error; 1111 bp->b_resid = (error ? bp->b_bcount : 0); 1112 biodone(bp); 1113 1114 mutex_enter(&sc->sc_slock); 1115 continue; 1116 } 1117 1118 /* 1119 * Clean the cache 1120 */ 1121 for (i = 0; i < sc->sc_cache_size; i++) { 1122 crotor = (crotor + 1) % sc->sc_cache_size; 1123 scp = sc->sc_cache + crotor; 1124 if (scp->fc_type != FSS_CACHE_VALID) 1125 continue; 1126 mutex_exit(&sc->sc_slock); 1127 1128 thread_idle = false; 1129 indirp = fss_bs_indir(sc, scp->fc_cluster); 1130 if (indirp != NULL) { 1131 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1132 0, FSS_CLSIZE(sc), scp->fc_data); 1133 } else 1134 error = EIO; 1135 1136 mutex_enter(&sc->sc_slock); 1137 if (error == 0) { 1138 *indirp = sc->sc_clnext++; 1139 sc->sc_indir_dirty = 1; 1140 } else 1141 fss_error(sc, "write error on backing store"); 1142 1143 scp->fc_type = FSS_CACHE_FREE; 1144 cv_signal(&sc->sc_cache_cv); 1145 break; 1146 } 1147 1148 /* 1149 * Process I/O requests 1150 */ 1151 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1152 continue; 1153 is_valid = FSS_ISVALID(sc); 1154 is_read = (bp->b_flags & B_READ); 1155 thread_idle = false; 1156 1157 if (!is_valid) { 1158 mutex_exit(&sc->sc_slock); 1159 1160 bp->b_error = ENXIO; 1161 bp->b_resid = bp->b_bcount; 1162 biodone(bp); 1163 1164 mutex_enter(&sc->sc_slock); 1165 continue; 1166 } 1167 1168 disk_busy(sc->sc_dkdev); 1169 1170 /* 1171 * First read from the snapshotted block device unless 1172 * this request is completely covered by backing store. 1173 */ 1174 1175 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1176 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1177 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1178 error = 0; 1179 bp->b_resid = 0; 1180 bp->b_error = 0; 1181 for (c = cl; c <= ch; c++) { 1182 if (isset(sc->sc_copied, c)) 1183 continue; 1184 mutex_exit(&sc->sc_slock); 1185 1186 /* Not on backing store, read from device. */ 1187 nbp = getiobuf(NULL, true); 1188 nbp->b_flags = B_READ; 1189 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1190 nbp->b_bufsize = bp->b_bcount; 1191 nbp->b_data = bp->b_data; 1192 nbp->b_blkno = bp->b_blkno; 1193 nbp->b_lblkno = 0; 1194 nbp->b_dev = sc->sc_bdev; 1195 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1196 1197 bdev_strategy(nbp); 1198 1199 error = biowait(nbp); 1200 if (error != 0) { 1201 bp->b_resid = bp->b_bcount; 1202 bp->b_error = nbp->b_error; 1203 disk_unbusy(sc->sc_dkdev, 0, is_read); 1204 biodone(bp); 1205 } 1206 putiobuf(nbp); 1207 1208 mutex_enter(&sc->sc_slock); 1209 break; 1210 } 1211 if (error) 1212 continue; 1213 1214 /* 1215 * Replace those parts that have been saved to backing store. 1216 */ 1217 1218 addr = bp->b_data; 1219 todo = bp->b_bcount; 1220 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1221 len = FSS_CLSIZE(sc)-off; 1222 if (len > todo) 1223 len = todo; 1224 if (isclr(sc->sc_copied, c)) 1225 continue; 1226 mutex_exit(&sc->sc_slock); 1227 1228 indirp = fss_bs_indir(sc, c); 1229 if (indirp == NULL || *indirp == 0) { 1230 /* 1231 * Not on backing store. Either in cache 1232 * or hole in the snapshotted block device. 1233 */ 1234 1235 mutex_enter(&sc->sc_slock); 1236 for (scp = sc->sc_cache; scp < scl; scp++) 1237 if (scp->fc_type == FSS_CACHE_VALID && 1238 scp->fc_cluster == c) 1239 break; 1240 if (scp < scl) 1241 memcpy(addr, (char *)scp->fc_data+off, 1242 len); 1243 else 1244 memset(addr, 0, len); 1245 continue; 1246 } 1247 1248 /* 1249 * Read from backing store. 1250 */ 1251 error = 1252 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1253 1254 mutex_enter(&sc->sc_slock); 1255 if (error) { 1256 bp->b_resid = bp->b_bcount; 1257 bp->b_error = error; 1258 break; 1259 } 1260 } 1261 mutex_exit(&sc->sc_slock); 1262 1263 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1264 biodone(bp); 1265 1266 mutex_enter(&sc->sc_slock); 1267 } 1268 } 1269 1270 #ifdef _MODULE 1271 1272 #include <sys/module.h> 1273 1274 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1275 CFDRIVER_DECL(fss, DV_DISK, NULL); 1276 1277 static int 1278 fss_modcmd(modcmd_t cmd, void *arg) 1279 { 1280 int bmajor = -1, cmajor = -1, error = 0; 1281 1282 switch (cmd) { 1283 case MODULE_CMD_INIT: 1284 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1285 error = config_cfdriver_attach(&fss_cd); 1286 if (error) { 1287 mutex_destroy(&fss_device_lock); 1288 break; 1289 } 1290 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1291 if (error) { 1292 config_cfdriver_detach(&fss_cd); 1293 mutex_destroy(&fss_device_lock); 1294 break; 1295 } 1296 error = devsw_attach(fss_cd.cd_name, 1297 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1298 if (error == EEXIST) 1299 error = 0; 1300 if (error) { 1301 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1302 config_cfdriver_detach(&fss_cd); 1303 mutex_destroy(&fss_device_lock); 1304 break; 1305 } 1306 break; 1307 1308 case MODULE_CMD_FINI: 1309 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1310 if (error) 1311 break; 1312 config_cfdriver_detach(&fss_cd); 1313 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1314 mutex_destroy(&fss_device_lock); 1315 break; 1316 1317 default: 1318 error = ENOTTY; 1319 break; 1320 } 1321 1322 return error; 1323 } 1324 1325 #endif /* _MODULE */ 1326