1 /* 2 * ioctl.c - NILFS ioctl operations. 3 * 4 * Copyright (C) 2007, 2008 Nippon Telegraph and Telephone Corporation. 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License as published by 8 * the Free Software Foundation; either version 2 of the License, or 9 * (at your option) any later version. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 19 * 20 * Written by Koji Sato <koji@osrg.net>. 21 */ 22 23 #include <linux/fs.h> 24 #include <linux/wait.h> 25 #include <linux/smp_lock.h> /* lock_kernel(), unlock_kernel() */ 26 #include <linux/capability.h> /* capable() */ 27 #include <linux/uaccess.h> /* copy_from_user(), copy_to_user() */ 28 #include <linux/vmalloc.h> 29 #include <linux/nilfs2_fs.h> 30 #include "nilfs.h" 31 #include "segment.h" 32 #include "bmap.h" 33 #include "cpfile.h" 34 #include "sufile.h" 35 #include "dat.h" 36 37 38 static int nilfs_ioctl_wrap_copy(struct the_nilfs *nilfs, 39 struct nilfs_argv *argv, int dir, 40 ssize_t (*dofunc)(struct the_nilfs *, 41 __u64 *, int, 42 void *, size_t, size_t)) 43 { 44 void *buf; 45 void __user *base = (void __user *)(unsigned long)argv->v_base; 46 size_t maxmembs, total, n; 47 ssize_t nr; 48 int ret, i; 49 __u64 pos, ppos; 50 51 if (argv->v_nmembs == 0) 52 return 0; 53 54 if (argv->v_size > PAGE_SIZE) 55 return -EINVAL; 56 57 buf = (void *)__get_free_pages(GFP_NOFS, 0); 58 if (unlikely(!buf)) 59 return -ENOMEM; 60 maxmembs = PAGE_SIZE / argv->v_size; 61 62 ret = 0; 63 total = 0; 64 pos = argv->v_index; 65 for (i = 0; i < argv->v_nmembs; i += n) { 66 n = (argv->v_nmembs - i < maxmembs) ? 67 argv->v_nmembs - i : maxmembs; 68 if ((dir & _IOC_WRITE) && 69 copy_from_user(buf, base + argv->v_size * i, 70 argv->v_size * n)) { 71 ret = -EFAULT; 72 break; 73 } 74 ppos = pos; 75 nr = dofunc(nilfs, &pos, argv->v_flags, buf, argv->v_size, 76 n); 77 if (nr < 0) { 78 ret = nr; 79 break; 80 } 81 if ((dir & _IOC_READ) && 82 copy_to_user(base + argv->v_size * i, buf, 83 argv->v_size * nr)) { 84 ret = -EFAULT; 85 break; 86 } 87 total += nr; 88 if ((size_t)nr < n) 89 break; 90 if (pos == ppos) 91 pos += n; 92 } 93 argv->v_nmembs = total; 94 95 free_pages((unsigned long)buf, 0); 96 return ret; 97 } 98 99 static int nilfs_ioctl_change_cpmode(struct inode *inode, struct file *filp, 100 unsigned int cmd, void __user *argp) 101 { 102 struct inode *cpfile = NILFS_SB(inode->i_sb)->s_nilfs->ns_cpfile; 103 struct nilfs_transaction_info ti; 104 struct nilfs_cpmode cpmode; 105 int ret; 106 107 if (!capable(CAP_SYS_ADMIN)) 108 return -EPERM; 109 if (copy_from_user(&cpmode, argp, sizeof(cpmode))) 110 return -EFAULT; 111 112 nilfs_transaction_begin(inode->i_sb, &ti, 0); 113 ret = nilfs_cpfile_change_cpmode( 114 cpfile, cpmode.cm_cno, cpmode.cm_mode); 115 if (unlikely(ret < 0)) { 116 nilfs_transaction_abort(inode->i_sb); 117 return ret; 118 } 119 nilfs_transaction_commit(inode->i_sb); /* never fails */ 120 return ret; 121 } 122 123 static int 124 nilfs_ioctl_delete_checkpoint(struct inode *inode, struct file *filp, 125 unsigned int cmd, void __user *argp) 126 { 127 struct inode *cpfile = NILFS_SB(inode->i_sb)->s_nilfs->ns_cpfile; 128 struct nilfs_transaction_info ti; 129 __u64 cno; 130 int ret; 131 132 if (!capable(CAP_SYS_ADMIN)) 133 return -EPERM; 134 if (copy_from_user(&cno, argp, sizeof(cno))) 135 return -EFAULT; 136 137 nilfs_transaction_begin(inode->i_sb, &ti, 0); 138 ret = nilfs_cpfile_delete_checkpoint(cpfile, cno); 139 if (unlikely(ret < 0)) { 140 nilfs_transaction_abort(inode->i_sb); 141 return ret; 142 } 143 nilfs_transaction_commit(inode->i_sb); /* never fails */ 144 return ret; 145 } 146 147 static ssize_t 148 nilfs_ioctl_do_get_cpinfo(struct the_nilfs *nilfs, __u64 *posp, int flags, 149 void *buf, size_t size, size_t nmembs) 150 { 151 int ret; 152 153 down_read(&nilfs->ns_segctor_sem); 154 ret = nilfs_cpfile_get_cpinfo(nilfs->ns_cpfile, posp, flags, buf, 155 size, nmembs); 156 up_read(&nilfs->ns_segctor_sem); 157 return ret; 158 } 159 160 static int nilfs_ioctl_get_cpstat(struct inode *inode, struct file *filp, 161 unsigned int cmd, void __user *argp) 162 { 163 struct the_nilfs *nilfs = NILFS_SB(inode->i_sb)->s_nilfs; 164 struct nilfs_cpstat cpstat; 165 int ret; 166 167 down_read(&nilfs->ns_segctor_sem); 168 ret = nilfs_cpfile_get_stat(nilfs->ns_cpfile, &cpstat); 169 up_read(&nilfs->ns_segctor_sem); 170 if (ret < 0) 171 return ret; 172 173 if (copy_to_user(argp, &cpstat, sizeof(cpstat))) 174 ret = -EFAULT; 175 return ret; 176 } 177 178 static ssize_t 179 nilfs_ioctl_do_get_suinfo(struct the_nilfs *nilfs, __u64 *posp, int flags, 180 void *buf, size_t size, size_t nmembs) 181 { 182 int ret; 183 184 down_read(&nilfs->ns_segctor_sem); 185 ret = nilfs_sufile_get_suinfo(nilfs->ns_sufile, *posp, buf, size, 186 nmembs); 187 up_read(&nilfs->ns_segctor_sem); 188 return ret; 189 } 190 191 static int nilfs_ioctl_get_sustat(struct inode *inode, struct file *filp, 192 unsigned int cmd, void __user *argp) 193 { 194 struct the_nilfs *nilfs = NILFS_SB(inode->i_sb)->s_nilfs; 195 struct nilfs_sustat sustat; 196 int ret; 197 198 down_read(&nilfs->ns_segctor_sem); 199 ret = nilfs_sufile_get_stat(nilfs->ns_sufile, &sustat); 200 up_read(&nilfs->ns_segctor_sem); 201 if (ret < 0) 202 return ret; 203 204 if (copy_to_user(argp, &sustat, sizeof(sustat))) 205 ret = -EFAULT; 206 return ret; 207 } 208 209 static ssize_t 210 nilfs_ioctl_do_get_vinfo(struct the_nilfs *nilfs, __u64 *posp, int flags, 211 void *buf, size_t size, size_t nmembs) 212 { 213 int ret; 214 215 down_read(&nilfs->ns_segctor_sem); 216 ret = nilfs_dat_get_vinfo(nilfs_dat_inode(nilfs), buf, size, nmembs); 217 up_read(&nilfs->ns_segctor_sem); 218 return ret; 219 } 220 221 static ssize_t 222 nilfs_ioctl_do_get_bdescs(struct the_nilfs *nilfs, __u64 *posp, int flags, 223 void *buf, size_t size, size_t nmembs) 224 { 225 struct inode *dat = nilfs_dat_inode(nilfs); 226 struct nilfs_bmap *bmap = NILFS_I(dat)->i_bmap; 227 struct nilfs_bdesc *bdescs = buf; 228 int ret, i; 229 230 down_read(&nilfs->ns_segctor_sem); 231 for (i = 0; i < nmembs; i++) { 232 ret = nilfs_bmap_lookup_at_level(bmap, 233 bdescs[i].bd_offset, 234 bdescs[i].bd_level + 1, 235 &bdescs[i].bd_blocknr); 236 if (ret < 0) { 237 if (ret != -ENOENT) { 238 up_read(&nilfs->ns_segctor_sem); 239 return ret; 240 } 241 bdescs[i].bd_blocknr = 0; 242 } 243 } 244 up_read(&nilfs->ns_segctor_sem); 245 return nmembs; 246 } 247 248 static int nilfs_ioctl_get_bdescs(struct inode *inode, struct file *filp, 249 unsigned int cmd, void __user *argp) 250 { 251 struct the_nilfs *nilfs = NILFS_SB(inode->i_sb)->s_nilfs; 252 struct nilfs_argv argv; 253 int ret; 254 255 if (copy_from_user(&argv, argp, sizeof(argv))) 256 return -EFAULT; 257 258 if (argv.v_size != sizeof(struct nilfs_bdesc)) 259 return -EINVAL; 260 261 ret = nilfs_ioctl_wrap_copy(nilfs, &argv, _IOC_DIR(cmd), 262 nilfs_ioctl_do_get_bdescs); 263 if (ret < 0) 264 return ret; 265 266 if (copy_to_user(argp, &argv, sizeof(argv))) 267 ret = -EFAULT; 268 return ret; 269 } 270 271 static int nilfs_ioctl_move_inode_block(struct inode *inode, 272 struct nilfs_vdesc *vdesc, 273 struct list_head *buffers) 274 { 275 struct buffer_head *bh; 276 int ret; 277 278 if (vdesc->vd_flags == 0) 279 ret = nilfs_gccache_submit_read_data( 280 inode, vdesc->vd_offset, vdesc->vd_blocknr, 281 vdesc->vd_vblocknr, &bh); 282 else 283 ret = nilfs_gccache_submit_read_node( 284 inode, vdesc->vd_blocknr, vdesc->vd_vblocknr, &bh); 285 286 if (unlikely(ret < 0)) { 287 if (ret == -ENOENT) 288 printk(KERN_CRIT 289 "%s: invalid virtual block address (%s): " 290 "ino=%llu, cno=%llu, offset=%llu, " 291 "blocknr=%llu, vblocknr=%llu\n", 292 __func__, vdesc->vd_flags ? "node" : "data", 293 (unsigned long long)vdesc->vd_ino, 294 (unsigned long long)vdesc->vd_cno, 295 (unsigned long long)vdesc->vd_offset, 296 (unsigned long long)vdesc->vd_blocknr, 297 (unsigned long long)vdesc->vd_vblocknr); 298 return ret; 299 } 300 bh->b_private = vdesc; 301 list_add_tail(&bh->b_assoc_buffers, buffers); 302 return 0; 303 } 304 305 static int nilfs_ioctl_move_blocks(struct the_nilfs *nilfs, 306 struct nilfs_argv *argv, void *buf) 307 { 308 size_t nmembs = argv->v_nmembs; 309 struct inode *inode; 310 struct nilfs_vdesc *vdesc; 311 struct buffer_head *bh, *n; 312 LIST_HEAD(buffers); 313 ino_t ino; 314 __u64 cno; 315 int i, ret; 316 317 for (i = 0, vdesc = buf; i < nmembs; ) { 318 ino = vdesc->vd_ino; 319 cno = vdesc->vd_cno; 320 inode = nilfs_gc_iget(nilfs, ino, cno); 321 if (unlikely(inode == NULL)) { 322 ret = -ENOMEM; 323 goto failed; 324 } 325 do { 326 ret = nilfs_ioctl_move_inode_block(inode, vdesc, 327 &buffers); 328 if (unlikely(ret < 0)) 329 goto failed; 330 vdesc++; 331 } while (++i < nmembs && 332 vdesc->vd_ino == ino && vdesc->vd_cno == cno); 333 } 334 335 list_for_each_entry_safe(bh, n, &buffers, b_assoc_buffers) { 336 ret = nilfs_gccache_wait_and_mark_dirty(bh); 337 if (unlikely(ret < 0)) { 338 if (ret == -EEXIST) { 339 vdesc = bh->b_private; 340 printk(KERN_CRIT 341 "%s: conflicting %s buffer: " 342 "ino=%llu, cno=%llu, offset=%llu, " 343 "blocknr=%llu, vblocknr=%llu\n", 344 __func__, 345 vdesc->vd_flags ? "node" : "data", 346 (unsigned long long)vdesc->vd_ino, 347 (unsigned long long)vdesc->vd_cno, 348 (unsigned long long)vdesc->vd_offset, 349 (unsigned long long)vdesc->vd_blocknr, 350 (unsigned long long)vdesc->vd_vblocknr); 351 } 352 goto failed; 353 } 354 list_del_init(&bh->b_assoc_buffers); 355 bh->b_private = NULL; 356 brelse(bh); 357 } 358 return nmembs; 359 360 failed: 361 list_for_each_entry_safe(bh, n, &buffers, b_assoc_buffers) { 362 list_del_init(&bh->b_assoc_buffers); 363 bh->b_private = NULL; 364 brelse(bh); 365 } 366 return ret; 367 } 368 369 static int nilfs_ioctl_delete_checkpoints(struct the_nilfs *nilfs, 370 struct nilfs_argv *argv, void *buf) 371 { 372 size_t nmembs = argv->v_nmembs; 373 struct inode *cpfile = nilfs->ns_cpfile; 374 struct nilfs_period *periods = buf; 375 int ret, i; 376 377 for (i = 0; i < nmembs; i++) { 378 ret = nilfs_cpfile_delete_checkpoints( 379 cpfile, periods[i].p_start, periods[i].p_end); 380 if (ret < 0) 381 return ret; 382 } 383 return nmembs; 384 } 385 386 static int nilfs_ioctl_free_vblocknrs(struct the_nilfs *nilfs, 387 struct nilfs_argv *argv, void *buf) 388 { 389 size_t nmembs = argv->v_nmembs; 390 int ret; 391 392 ret = nilfs_dat_freev(nilfs_dat_inode(nilfs), buf, nmembs); 393 394 return (ret < 0) ? ret : nmembs; 395 } 396 397 static int nilfs_ioctl_mark_blocks_dirty(struct the_nilfs *nilfs, 398 struct nilfs_argv *argv, void *buf) 399 { 400 size_t nmembs = argv->v_nmembs; 401 struct inode *dat = nilfs_dat_inode(nilfs); 402 struct nilfs_bmap *bmap = NILFS_I(dat)->i_bmap; 403 struct nilfs_bdesc *bdescs = buf; 404 int ret, i; 405 406 for (i = 0; i < nmembs; i++) { 407 /* XXX: use macro or inline func to check liveness */ 408 ret = nilfs_bmap_lookup_at_level(bmap, 409 bdescs[i].bd_offset, 410 bdescs[i].bd_level + 1, 411 &bdescs[i].bd_blocknr); 412 if (ret < 0) { 413 if (ret != -ENOENT) 414 return ret; 415 bdescs[i].bd_blocknr = 0; 416 } 417 if (bdescs[i].bd_blocknr != bdescs[i].bd_oblocknr) 418 /* skip dead block */ 419 continue; 420 if (bdescs[i].bd_level == 0) { 421 ret = nilfs_mdt_mark_block_dirty(dat, 422 bdescs[i].bd_offset); 423 if (ret < 0) { 424 WARN_ON(ret == -ENOENT); 425 return ret; 426 } 427 } else { 428 ret = nilfs_bmap_mark(bmap, bdescs[i].bd_offset, 429 bdescs[i].bd_level); 430 if (ret < 0) { 431 WARN_ON(ret == -ENOENT); 432 return ret; 433 } 434 } 435 } 436 return nmembs; 437 } 438 439 int nilfs_ioctl_prepare_clean_segments(struct the_nilfs *nilfs, 440 struct nilfs_argv *argv, void **kbufs) 441 { 442 const char *msg; 443 int ret; 444 445 ret = nilfs_ioctl_delete_checkpoints(nilfs, &argv[1], kbufs[1]); 446 if (ret < 0) { 447 /* 448 * can safely abort because checkpoints can be removed 449 * independently. 450 */ 451 msg = "cannot delete checkpoints"; 452 goto failed; 453 } 454 ret = nilfs_ioctl_free_vblocknrs(nilfs, &argv[2], kbufs[2]); 455 if (ret < 0) { 456 /* 457 * can safely abort because DAT file is updated atomically 458 * using a copy-on-write technique. 459 */ 460 msg = "cannot delete virtual blocks from DAT file"; 461 goto failed; 462 } 463 ret = nilfs_ioctl_mark_blocks_dirty(nilfs, &argv[3], kbufs[3]); 464 if (ret < 0) { 465 /* 466 * can safely abort because the operation is nondestructive. 467 */ 468 msg = "cannot mark copying blocks dirty"; 469 goto failed; 470 } 471 return 0; 472 473 failed: 474 nilfs_remove_all_gcinode(nilfs); 475 printk(KERN_ERR "NILFS: GC failed during preparation: %s: err=%d\n", 476 msg, ret); 477 return ret; 478 } 479 480 static int nilfs_ioctl_clean_segments(struct inode *inode, struct file *filp, 481 unsigned int cmd, void __user *argp) 482 { 483 struct nilfs_argv argv[5]; 484 const static size_t argsz[5] = { 485 sizeof(struct nilfs_vdesc), 486 sizeof(struct nilfs_period), 487 sizeof(__u64), 488 sizeof(struct nilfs_bdesc), 489 sizeof(__u64), 490 }; 491 void __user *base; 492 void *kbufs[5]; 493 struct the_nilfs *nilfs; 494 size_t len, nsegs; 495 int n, ret; 496 497 if (!capable(CAP_SYS_ADMIN)) 498 return -EPERM; 499 500 if (copy_from_user(argv, argp, sizeof(argv))) 501 return -EFAULT; 502 503 nsegs = argv[4].v_nmembs; 504 if (argv[4].v_size != argsz[4]) 505 return -EINVAL; 506 /* 507 * argv[4] points to segment numbers this ioctl cleans. We 508 * use kmalloc() for its buffer because memory used for the 509 * segment numbers is enough small. 510 */ 511 kbufs[4] = memdup_user((void __user *)(unsigned long)argv[4].v_base, 512 nsegs * sizeof(__u64)); 513 if (IS_ERR(kbufs[4])) 514 return PTR_ERR(kbufs[4]); 515 516 nilfs = NILFS_SB(inode->i_sb)->s_nilfs; 517 518 for (n = 0; n < 4; n++) { 519 ret = -EINVAL; 520 if (argv[n].v_size != argsz[n]) 521 goto out_free; 522 523 if (argv[n].v_nmembs > nsegs * nilfs->ns_blocks_per_segment) 524 goto out_free; 525 526 len = argv[n].v_size * argv[n].v_nmembs; 527 base = (void __user *)(unsigned long)argv[n].v_base; 528 if (len == 0) { 529 kbufs[n] = NULL; 530 continue; 531 } 532 533 kbufs[n] = vmalloc(len); 534 if (!kbufs[n]) { 535 ret = -ENOMEM; 536 goto out_free; 537 } 538 if (copy_from_user(kbufs[n], base, len)) { 539 ret = -EFAULT; 540 vfree(kbufs[n]); 541 goto out_free; 542 } 543 } 544 545 /* 546 * nilfs_ioctl_move_blocks() will call nilfs_gc_iget(), 547 * which will operates an inode list without blocking. 548 * To protect the list from concurrent operations, 549 * nilfs_ioctl_move_blocks should be atomic operation. 550 */ 551 if (test_and_set_bit(THE_NILFS_GC_RUNNING, &nilfs->ns_flags)) { 552 ret = -EBUSY; 553 goto out_free; 554 } 555 556 ret = nilfs_ioctl_move_blocks(nilfs, &argv[0], kbufs[0]); 557 if (ret < 0) 558 printk(KERN_ERR "NILFS: GC failed during preparation: " 559 "cannot read source blocks: err=%d\n", ret); 560 else 561 ret = nilfs_clean_segments(inode->i_sb, argv, kbufs); 562 563 clear_nilfs_gc_running(nilfs); 564 565 out_free: 566 while (--n >= 0) 567 vfree(kbufs[n]); 568 kfree(kbufs[4]); 569 return ret; 570 } 571 572 static int nilfs_ioctl_sync(struct inode *inode, struct file *filp, 573 unsigned int cmd, void __user *argp) 574 { 575 __u64 cno; 576 int ret; 577 578 ret = nilfs_construct_segment(inode->i_sb); 579 if (ret < 0) 580 return ret; 581 582 if (argp != NULL) { 583 cno = NILFS_SB(inode->i_sb)->s_nilfs->ns_cno - 1; 584 if (copy_to_user(argp, &cno, sizeof(cno))) 585 return -EFAULT; 586 } 587 return 0; 588 } 589 590 static int nilfs_ioctl_get_info(struct inode *inode, struct file *filp, 591 unsigned int cmd, void __user *argp, 592 size_t membsz, 593 ssize_t (*dofunc)(struct the_nilfs *, 594 __u64 *, int, 595 void *, size_t, size_t)) 596 597 { 598 struct the_nilfs *nilfs = NILFS_SB(inode->i_sb)->s_nilfs; 599 struct nilfs_argv argv; 600 int ret; 601 602 if (copy_from_user(&argv, argp, sizeof(argv))) 603 return -EFAULT; 604 605 if (argv.v_size < membsz) 606 return -EINVAL; 607 608 ret = nilfs_ioctl_wrap_copy(nilfs, &argv, _IOC_DIR(cmd), dofunc); 609 if (ret < 0) 610 return ret; 611 612 if (copy_to_user(argp, &argv, sizeof(argv))) 613 ret = -EFAULT; 614 return ret; 615 } 616 617 long nilfs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 618 { 619 struct inode *inode = filp->f_dentry->d_inode; 620 void __user *argp = (void * __user *)arg; 621 622 switch (cmd) { 623 case NILFS_IOCTL_CHANGE_CPMODE: 624 return nilfs_ioctl_change_cpmode(inode, filp, cmd, argp); 625 case NILFS_IOCTL_DELETE_CHECKPOINT: 626 return nilfs_ioctl_delete_checkpoint(inode, filp, cmd, argp); 627 case NILFS_IOCTL_GET_CPINFO: 628 return nilfs_ioctl_get_info(inode, filp, cmd, argp, 629 sizeof(struct nilfs_cpinfo), 630 nilfs_ioctl_do_get_cpinfo); 631 case NILFS_IOCTL_GET_CPSTAT: 632 return nilfs_ioctl_get_cpstat(inode, filp, cmd, argp); 633 case NILFS_IOCTL_GET_SUINFO: 634 return nilfs_ioctl_get_info(inode, filp, cmd, argp, 635 sizeof(struct nilfs_suinfo), 636 nilfs_ioctl_do_get_suinfo); 637 case NILFS_IOCTL_GET_SUSTAT: 638 return nilfs_ioctl_get_sustat(inode, filp, cmd, argp); 639 case NILFS_IOCTL_GET_VINFO: 640 return nilfs_ioctl_get_info(inode, filp, cmd, argp, 641 sizeof(struct nilfs_vinfo), 642 nilfs_ioctl_do_get_vinfo); 643 case NILFS_IOCTL_GET_BDESCS: 644 return nilfs_ioctl_get_bdescs(inode, filp, cmd, argp); 645 case NILFS_IOCTL_CLEAN_SEGMENTS: 646 return nilfs_ioctl_clean_segments(inode, filp, cmd, argp); 647 case NILFS_IOCTL_SYNC: 648 return nilfs_ioctl_sync(inode, filp, cmd, argp); 649 default: 650 return -ENOTTY; 651 } 652 } 653