1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. 4 * 5 * This copyrighted material is made available to anyone wishing to use, 6 * modify, copy, or redistribute it subject to the terms and conditions 7 * of the GNU General Public License version 2. 8 */ 9 10 #include <linux/sched.h> 11 #include <linux/slab.h> 12 #include <linux/spinlock.h> 13 #include <linux/completion.h> 14 #include <linux/buffer_head.h> 15 #include <linux/posix_acl.h> 16 #include <linux/sort.h> 17 #include <linux/gfs2_ondisk.h> 18 #include <linux/crc32.h> 19 #include <linux/lm_interface.h> 20 #include <linux/security.h> 21 22 #include "gfs2.h" 23 #include "incore.h" 24 #include "acl.h" 25 #include "bmap.h" 26 #include "dir.h" 27 #include "eattr.h" 28 #include "glock.h" 29 #include "glops.h" 30 #include "inode.h" 31 #include "log.h" 32 #include "meta_io.h" 33 #include "ops_address.h" 34 #include "ops_file.h" 35 #include "ops_inode.h" 36 #include "quota.h" 37 #include "rgrp.h" 38 #include "trans.h" 39 #include "util.h" 40 41 struct gfs2_inum_range_host { 42 u64 ir_start; 43 u64 ir_length; 44 }; 45 46 static int iget_test(struct inode *inode, void *opaque) 47 { 48 struct gfs2_inode *ip = GFS2_I(inode); 49 u64 *no_addr = opaque; 50 51 if (ip->i_no_addr == *no_addr && 52 inode->i_private != NULL) 53 return 1; 54 55 return 0; 56 } 57 58 static int iget_set(struct inode *inode, void *opaque) 59 { 60 struct gfs2_inode *ip = GFS2_I(inode); 61 u64 *no_addr = opaque; 62 63 inode->i_ino = (unsigned long)*no_addr; 64 ip->i_no_addr = *no_addr; 65 return 0; 66 } 67 68 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr) 69 { 70 unsigned long hash = (unsigned long)no_addr; 71 return ilookup5(sb, hash, iget_test, &no_addr); 72 } 73 74 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr) 75 { 76 unsigned long hash = (unsigned long)no_addr; 77 return iget5_locked(sb, hash, iget_test, iget_set, &no_addr); 78 } 79 80 /** 81 * gfs2_inode_lookup - Lookup an inode 82 * @sb: The super block 83 * @no_addr: The inode number 84 * @type: The type of the inode 85 * 86 * Returns: A VFS inode, or an error 87 */ 88 89 struct inode *gfs2_inode_lookup(struct super_block *sb, 90 unsigned int type, 91 u64 no_addr, 92 u64 no_formal_ino) 93 { 94 struct inode *inode = gfs2_iget(sb, no_addr); 95 struct gfs2_inode *ip = GFS2_I(inode); 96 struct gfs2_glock *io_gl; 97 int error; 98 99 if (!inode) 100 return ERR_PTR(-ENOBUFS); 101 102 if (inode->i_state & I_NEW) { 103 struct gfs2_sbd *sdp = GFS2_SB(inode); 104 umode_t mode; 105 inode->i_private = ip; 106 ip->i_no_formal_ino = no_formal_ino; 107 108 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl); 109 if (unlikely(error)) 110 goto fail; 111 ip->i_gl->gl_object = ip; 112 113 error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl); 114 if (unlikely(error)) 115 goto fail_put; 116 117 set_bit(GIF_INVALID, &ip->i_flags); 118 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh); 119 if (unlikely(error)) 120 goto fail_iopen; 121 ip->i_iopen_gh.gh_gl->gl_object = ip; 122 123 gfs2_glock_put(io_gl); 124 125 /* 126 * We must read the inode in order to work out its type in 127 * this case. Note that this doesn't happen often as we normally 128 * know the type beforehand. This code path only occurs during 129 * unlinked inode recovery (where it is safe to do this glock, 130 * which is not true in the general case). 131 */ 132 inode->i_mode = mode = DT2IF(type); 133 if (type == DT_UNKNOWN) { 134 struct gfs2_holder gh; 135 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 136 if (unlikely(error)) 137 goto fail_glock; 138 /* Inode is now uptodate */ 139 mode = inode->i_mode; 140 gfs2_glock_dq_uninit(&gh); 141 } 142 143 if (S_ISREG(mode)) { 144 inode->i_op = &gfs2_file_iops; 145 inode->i_fop = &gfs2_file_fops; 146 inode->i_mapping->a_ops = &gfs2_file_aops; 147 } else if (S_ISDIR(mode)) { 148 inode->i_op = &gfs2_dir_iops; 149 inode->i_fop = &gfs2_dir_fops; 150 } else if (S_ISLNK(mode)) { 151 inode->i_op = &gfs2_symlink_iops; 152 } else { 153 inode->i_op = &gfs2_dev_iops; 154 } 155 156 unlock_new_inode(inode); 157 } 158 159 return inode; 160 fail_glock: 161 gfs2_glock_dq(&ip->i_iopen_gh); 162 fail_iopen: 163 gfs2_glock_put(io_gl); 164 fail_put: 165 ip->i_gl->gl_object = NULL; 166 gfs2_glock_put(ip->i_gl); 167 fail: 168 iput(inode); 169 return ERR_PTR(error); 170 } 171 172 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf) 173 { 174 struct gfs2_dinode_host *di = &ip->i_di; 175 const struct gfs2_dinode *str = buf; 176 177 if (ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)) { 178 if (gfs2_consist_inode(ip)) 179 gfs2_dinode_print(ip); 180 return -EIO; 181 } 182 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino); 183 ip->i_inode.i_mode = be32_to_cpu(str->di_mode); 184 ip->i_inode.i_rdev = 0; 185 switch (ip->i_inode.i_mode & S_IFMT) { 186 case S_IFBLK: 187 case S_IFCHR: 188 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major), 189 be32_to_cpu(str->di_minor)); 190 break; 191 }; 192 193 ip->i_inode.i_uid = be32_to_cpu(str->di_uid); 194 ip->i_inode.i_gid = be32_to_cpu(str->di_gid); 195 /* 196 * We will need to review setting the nlink count here in the 197 * light of the forthcoming ro bind mount work. This is a reminder 198 * to do that. 199 */ 200 ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink); 201 di->di_size = be64_to_cpu(str->di_size); 202 i_size_write(&ip->i_inode, di->di_size); 203 di->di_blocks = be64_to_cpu(str->di_blocks); 204 gfs2_set_inode_blocks(&ip->i_inode); 205 ip->i_inode.i_atime.tv_sec = be64_to_cpu(str->di_atime); 206 ip->i_inode.i_atime.tv_nsec = be32_to_cpu(str->di_atime_nsec); 207 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime); 208 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec); 209 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime); 210 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec); 211 212 di->di_goal_meta = be64_to_cpu(str->di_goal_meta); 213 di->di_goal_data = be64_to_cpu(str->di_goal_data); 214 di->di_generation = be64_to_cpu(str->di_generation); 215 216 di->di_flags = be32_to_cpu(str->di_flags); 217 gfs2_set_inode_flags(&ip->i_inode); 218 di->di_height = be16_to_cpu(str->di_height); 219 220 di->di_depth = be16_to_cpu(str->di_depth); 221 di->di_entries = be32_to_cpu(str->di_entries); 222 223 di->di_eattr = be64_to_cpu(str->di_eattr); 224 return 0; 225 } 226 227 /** 228 * gfs2_inode_refresh - Refresh the incore copy of the dinode 229 * @ip: The GFS2 inode 230 * 231 * Returns: errno 232 */ 233 234 int gfs2_inode_refresh(struct gfs2_inode *ip) 235 { 236 struct buffer_head *dibh; 237 int error; 238 239 error = gfs2_meta_inode_buffer(ip, &dibh); 240 if (error) 241 return error; 242 243 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) { 244 brelse(dibh); 245 return -EIO; 246 } 247 248 error = gfs2_dinode_in(ip, dibh->b_data); 249 brelse(dibh); 250 clear_bit(GIF_INVALID, &ip->i_flags); 251 252 return error; 253 } 254 255 int gfs2_dinode_dealloc(struct gfs2_inode *ip) 256 { 257 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 258 struct gfs2_alloc *al; 259 struct gfs2_rgrpd *rgd; 260 int error; 261 262 if (ip->i_di.di_blocks != 1) { 263 if (gfs2_consist_inode(ip)) 264 gfs2_dinode_print(ip); 265 return -EIO; 266 } 267 268 al = gfs2_alloc_get(ip); 269 270 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 271 if (error) 272 goto out; 273 274 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 275 if (error) 276 goto out_qs; 277 278 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr); 279 if (!rgd) { 280 gfs2_consist_inode(ip); 281 error = -EIO; 282 goto out_rindex_relse; 283 } 284 285 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 286 &al->al_rgd_gh); 287 if (error) 288 goto out_rindex_relse; 289 290 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1); 291 if (error) 292 goto out_rg_gunlock; 293 294 gfs2_trans_add_gl(ip->i_gl); 295 296 gfs2_free_di(rgd, ip); 297 298 gfs2_trans_end(sdp); 299 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 300 301 out_rg_gunlock: 302 gfs2_glock_dq_uninit(&al->al_rgd_gh); 303 out_rindex_relse: 304 gfs2_glock_dq_uninit(&al->al_ri_gh); 305 out_qs: 306 gfs2_quota_unhold(ip); 307 out: 308 gfs2_alloc_put(ip); 309 return error; 310 } 311 312 /** 313 * gfs2_change_nlink - Change nlink count on inode 314 * @ip: The GFS2 inode 315 * @diff: The change in the nlink count required 316 * 317 * Returns: errno 318 */ 319 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 320 { 321 struct buffer_head *dibh; 322 u32 nlink; 323 int error; 324 325 BUG_ON(diff != 1 && diff != -1); 326 nlink = ip->i_inode.i_nlink + diff; 327 328 /* If we are reducing the nlink count, but the new value ends up being 329 bigger than the old one, we must have underflowed. */ 330 if (diff < 0 && nlink > ip->i_inode.i_nlink) { 331 if (gfs2_consist_inode(ip)) 332 gfs2_dinode_print(ip); 333 return -EIO; 334 } 335 336 error = gfs2_meta_inode_buffer(ip, &dibh); 337 if (error) 338 return error; 339 340 if (diff > 0) 341 inc_nlink(&ip->i_inode); 342 else 343 drop_nlink(&ip->i_inode); 344 345 ip->i_inode.i_ctime = CURRENT_TIME; 346 347 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 348 gfs2_dinode_out(ip, dibh->b_data); 349 brelse(dibh); 350 mark_inode_dirty(&ip->i_inode); 351 352 if (ip->i_inode.i_nlink == 0) 353 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 354 355 return error; 356 } 357 358 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 359 { 360 struct qstr qstr; 361 struct inode *inode; 362 gfs2_str2qstr(&qstr, name); 363 inode = gfs2_lookupi(dip, &qstr, 1, NULL); 364 /* gfs2_lookupi has inconsistent callers: vfs 365 * related routines expect NULL for no entry found, 366 * gfs2_lookup_simple callers expect ENOENT 367 * and do not check for NULL. 368 */ 369 if (inode == NULL) 370 return ERR_PTR(-ENOENT); 371 else 372 return inode; 373 } 374 375 376 /** 377 * gfs2_lookupi - Look up a filename in a directory and return its inode 378 * @d_gh: An initialized holder for the directory glock 379 * @name: The name of the inode to look for 380 * @is_root: If 1, ignore the caller's permissions 381 * @i_gh: An uninitialized holder for the new inode glock 382 * 383 * This can be called via the VFS filldir function when NFS is doing 384 * a readdirplus and the inode which its intending to stat isn't 385 * already in cache. In this case we must not take the directory glock 386 * again, since the readdir call will have already taken that lock. 387 * 388 * Returns: errno 389 */ 390 391 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 392 int is_root, struct nameidata *nd) 393 { 394 struct super_block *sb = dir->i_sb; 395 struct gfs2_inode *dip = GFS2_I(dir); 396 struct gfs2_holder d_gh; 397 int error = 0; 398 struct inode *inode = NULL; 399 int unlock = 0; 400 401 if (!name->len || name->len > GFS2_FNAMESIZE) 402 return ERR_PTR(-ENAMETOOLONG); 403 404 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 405 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 406 dir == sb->s_root->d_inode)) { 407 igrab(dir); 408 return dir; 409 } 410 411 if (gfs2_glock_is_locked_by_me(dip->i_gl) == 0) { 412 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 413 if (error) 414 return ERR_PTR(error); 415 unlock = 1; 416 } 417 418 if (!is_root) { 419 error = permission(dir, MAY_EXEC, NULL); 420 if (error) 421 goto out; 422 } 423 424 inode = gfs2_dir_search(dir, name); 425 if (IS_ERR(inode)) 426 error = PTR_ERR(inode); 427 out: 428 if (unlock) 429 gfs2_glock_dq_uninit(&d_gh); 430 if (error == -ENOENT) 431 return NULL; 432 return inode ? inode : ERR_PTR(error); 433 } 434 435 static void gfs2_inum_range_in(struct gfs2_inum_range_host *ir, const void *buf) 436 { 437 const struct gfs2_inum_range *str = buf; 438 439 ir->ir_start = be64_to_cpu(str->ir_start); 440 ir->ir_length = be64_to_cpu(str->ir_length); 441 } 442 443 static void gfs2_inum_range_out(const struct gfs2_inum_range_host *ir, void *buf) 444 { 445 struct gfs2_inum_range *str = buf; 446 447 str->ir_start = cpu_to_be64(ir->ir_start); 448 str->ir_length = cpu_to_be64(ir->ir_length); 449 } 450 451 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino) 452 { 453 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 454 struct buffer_head *bh; 455 struct gfs2_inum_range_host ir; 456 int error; 457 458 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 459 if (error) 460 return error; 461 mutex_lock(&sdp->sd_inum_mutex); 462 463 error = gfs2_meta_inode_buffer(ip, &bh); 464 if (error) { 465 mutex_unlock(&sdp->sd_inum_mutex); 466 gfs2_trans_end(sdp); 467 return error; 468 } 469 470 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 471 472 if (ir.ir_length) { 473 *formal_ino = ir.ir_start++; 474 ir.ir_length--; 475 gfs2_trans_add_bh(ip->i_gl, bh, 1); 476 gfs2_inum_range_out(&ir, 477 bh->b_data + sizeof(struct gfs2_dinode)); 478 brelse(bh); 479 mutex_unlock(&sdp->sd_inum_mutex); 480 gfs2_trans_end(sdp); 481 return 0; 482 } 483 484 brelse(bh); 485 486 mutex_unlock(&sdp->sd_inum_mutex); 487 gfs2_trans_end(sdp); 488 489 return 1; 490 } 491 492 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino) 493 { 494 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 495 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 496 struct gfs2_holder gh; 497 struct buffer_head *bh; 498 struct gfs2_inum_range_host ir; 499 int error; 500 501 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 502 if (error) 503 return error; 504 505 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 506 if (error) 507 goto out; 508 mutex_lock(&sdp->sd_inum_mutex); 509 510 error = gfs2_meta_inode_buffer(ip, &bh); 511 if (error) 512 goto out_end_trans; 513 514 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 515 516 if (!ir.ir_length) { 517 struct buffer_head *m_bh; 518 u64 x, y; 519 __be64 z; 520 521 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 522 if (error) 523 goto out_brelse; 524 525 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 526 x = y = be64_to_cpu(z); 527 ir.ir_start = x; 528 ir.ir_length = GFS2_INUM_QUANTUM; 529 x += GFS2_INUM_QUANTUM; 530 if (x < y) 531 gfs2_consist_inode(m_ip); 532 z = cpu_to_be64(x); 533 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 534 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z; 535 536 brelse(m_bh); 537 } 538 539 *formal_ino = ir.ir_start++; 540 ir.ir_length--; 541 542 gfs2_trans_add_bh(ip->i_gl, bh, 1); 543 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 544 545 out_brelse: 546 brelse(bh); 547 out_end_trans: 548 mutex_unlock(&sdp->sd_inum_mutex); 549 gfs2_trans_end(sdp); 550 out: 551 gfs2_glock_dq_uninit(&gh); 552 return error; 553 } 554 555 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum) 556 { 557 int error; 558 559 error = pick_formal_ino_1(sdp, inum); 560 if (error <= 0) 561 return error; 562 563 error = pick_formal_ino_2(sdp, inum); 564 565 return error; 566 } 567 568 /** 569 * create_ok - OK to create a new on-disk inode here? 570 * @dip: Directory in which dinode is to be created 571 * @name: Name of new dinode 572 * @mode: 573 * 574 * Returns: errno 575 */ 576 577 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 578 unsigned int mode) 579 { 580 int error; 581 582 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 583 if (error) 584 return error; 585 586 /* Don't create entries in an unlinked directory */ 587 if (!dip->i_inode.i_nlink) 588 return -EPERM; 589 590 error = gfs2_dir_check(&dip->i_inode, name, NULL); 591 switch (error) { 592 case -ENOENT: 593 error = 0; 594 break; 595 case 0: 596 return -EEXIST; 597 default: 598 return error; 599 } 600 601 if (dip->i_di.di_entries == (u32)-1) 602 return -EFBIG; 603 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1) 604 return -EMLINK; 605 606 return 0; 607 } 608 609 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 610 unsigned int *uid, unsigned int *gid) 611 { 612 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 613 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) { 614 if (S_ISDIR(*mode)) 615 *mode |= S_ISUID; 616 else if (dip->i_inode.i_uid != current->fsuid) 617 *mode &= ~07111; 618 *uid = dip->i_inode.i_uid; 619 } else 620 *uid = current->fsuid; 621 622 if (dip->i_inode.i_mode & S_ISGID) { 623 if (S_ISDIR(*mode)) 624 *mode |= S_ISGID; 625 *gid = dip->i_inode.i_gid; 626 } else 627 *gid = current->fsgid; 628 } 629 630 static int alloc_dinode(struct gfs2_inode *dip, u64 *no_addr, u64 *generation) 631 { 632 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 633 int error; 634 635 gfs2_alloc_get(dip); 636 637 dip->i_alloc.al_requested = RES_DINODE; 638 error = gfs2_inplace_reserve(dip); 639 if (error) 640 goto out; 641 642 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 643 if (error) 644 goto out_ipreserv; 645 646 *no_addr = gfs2_alloc_di(dip, generation); 647 648 gfs2_trans_end(sdp); 649 650 out_ipreserv: 651 gfs2_inplace_release(dip); 652 out: 653 gfs2_alloc_put(dip); 654 return error; 655 } 656 657 /** 658 * init_dinode - Fill in a new dinode structure 659 * @dip: the directory this inode is being created in 660 * @gl: The glock covering the new inode 661 * @inum: the inode number 662 * @mode: the file permissions 663 * @uid: 664 * @gid: 665 * 666 */ 667 668 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 669 const struct gfs2_inum_host *inum, unsigned int mode, 670 unsigned int uid, unsigned int gid, 671 const u64 *generation, dev_t dev) 672 { 673 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 674 struct gfs2_dinode *di; 675 struct buffer_head *dibh; 676 struct timespec tv = CURRENT_TIME; 677 678 dibh = gfs2_meta_new(gl, inum->no_addr); 679 gfs2_trans_add_bh(gl, dibh, 1); 680 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 681 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 682 di = (struct gfs2_dinode *)dibh->b_data; 683 684 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 685 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 686 di->di_mode = cpu_to_be32(mode); 687 di->di_uid = cpu_to_be32(uid); 688 di->di_gid = cpu_to_be32(gid); 689 di->di_nlink = 0; 690 di->di_size = 0; 691 di->di_blocks = cpu_to_be64(1); 692 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec); 693 di->di_major = cpu_to_be32(MAJOR(dev)); 694 di->di_minor = cpu_to_be32(MINOR(dev)); 695 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 696 di->di_generation = cpu_to_be64(*generation); 697 di->di_flags = 0; 698 699 if (S_ISREG(mode)) { 700 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 701 gfs2_tune_get(sdp, gt_new_files_jdata)) 702 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 703 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 704 gfs2_tune_get(sdp, gt_new_files_directio)) 705 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 706 } else if (S_ISDIR(mode)) { 707 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 708 GFS2_DIF_INHERIT_DIRECTIO); 709 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 710 GFS2_DIF_INHERIT_JDATA); 711 } 712 713 di->__pad1 = 0; 714 di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0); 715 di->di_height = 0; 716 di->__pad2 = 0; 717 di->__pad3 = 0; 718 di->di_depth = 0; 719 di->di_entries = 0; 720 memset(&di->__pad4, 0, sizeof(di->__pad4)); 721 di->di_eattr = 0; 722 di->di_atime_nsec = cpu_to_be32(tv.tv_nsec); 723 di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec); 724 di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec); 725 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 726 727 brelse(dibh); 728 } 729 730 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 731 unsigned int mode, const struct gfs2_inum_host *inum, 732 const u64 *generation, dev_t dev) 733 { 734 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 735 unsigned int uid, gid; 736 int error; 737 738 munge_mode_uid_gid(dip, &mode, &uid, &gid); 739 gfs2_alloc_get(dip); 740 741 error = gfs2_quota_lock(dip, uid, gid); 742 if (error) 743 goto out; 744 745 error = gfs2_quota_check(dip, uid, gid); 746 if (error) 747 goto out_quota; 748 749 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0); 750 if (error) 751 goto out_quota; 752 753 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev); 754 gfs2_quota_change(dip, +1, uid, gid); 755 gfs2_trans_end(sdp); 756 757 out_quota: 758 gfs2_quota_unlock(dip); 759 out: 760 gfs2_alloc_put(dip); 761 return error; 762 } 763 764 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name, 765 struct gfs2_inode *ip) 766 { 767 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 768 struct gfs2_alloc *al; 769 int alloc_required; 770 struct buffer_head *dibh; 771 int error; 772 773 al = gfs2_alloc_get(dip); 774 775 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 776 if (error) 777 goto fail; 778 779 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name); 780 if (alloc_required < 0) 781 goto fail; 782 if (alloc_required) { 783 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid); 784 if (error) 785 goto fail_quota_locks; 786 787 al->al_requested = sdp->sd_max_dirres; 788 789 error = gfs2_inplace_reserve(dip); 790 if (error) 791 goto fail_quota_locks; 792 793 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 794 al->al_rgd->rd_length + 795 2 * RES_DINODE + 796 RES_STATFS + RES_QUOTA, 0); 797 if (error) 798 goto fail_ipreserv; 799 } else { 800 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 801 if (error) 802 goto fail_quota_locks; 803 } 804 805 error = gfs2_dir_add(&dip->i_inode, name, ip, IF2DT(ip->i_inode.i_mode)); 806 if (error) 807 goto fail_end_trans; 808 809 error = gfs2_meta_inode_buffer(ip, &dibh); 810 if (error) 811 goto fail_end_trans; 812 ip->i_inode.i_nlink = 1; 813 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 814 gfs2_dinode_out(ip, dibh->b_data); 815 brelse(dibh); 816 return 0; 817 818 fail_end_trans: 819 gfs2_trans_end(sdp); 820 821 fail_ipreserv: 822 if (dip->i_alloc.al_rgd) 823 gfs2_inplace_release(dip); 824 825 fail_quota_locks: 826 gfs2_quota_unlock(dip); 827 828 fail: 829 gfs2_alloc_put(dip); 830 return error; 831 } 832 833 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip) 834 { 835 int err; 836 size_t len; 837 void *value; 838 char *name; 839 struct gfs2_ea_request er; 840 841 err = security_inode_init_security(&ip->i_inode, &dip->i_inode, 842 &name, &value, &len); 843 844 if (err) { 845 if (err == -EOPNOTSUPP) 846 return 0; 847 return err; 848 } 849 850 memset(&er, 0, sizeof(struct gfs2_ea_request)); 851 852 er.er_type = GFS2_EATYPE_SECURITY; 853 er.er_name = name; 854 er.er_data = value; 855 er.er_name_len = strlen(name); 856 er.er_data_len = len; 857 858 err = gfs2_ea_set_i(ip, &er); 859 860 kfree(value); 861 kfree(name); 862 863 return err; 864 } 865 866 /** 867 * gfs2_createi - Create a new inode 868 * @ghs: An array of two holders 869 * @name: The name of the new file 870 * @mode: the permissions on the new inode 871 * 872 * @ghs[0] is an initialized holder for the directory 873 * @ghs[1] is the holder for the inode lock 874 * 875 * If the return value is not NULL, the glocks on both the directory and the new 876 * file are held. A transaction has been started and an inplace reservation 877 * is held, as well. 878 * 879 * Returns: An inode 880 */ 881 882 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 883 unsigned int mode, dev_t dev) 884 { 885 struct inode *inode = NULL; 886 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 887 struct inode *dir = &dip->i_inode; 888 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 889 struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 }; 890 int error; 891 u64 generation; 892 893 if (!name->len || name->len > GFS2_FNAMESIZE) 894 return ERR_PTR(-ENAMETOOLONG); 895 896 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 897 error = gfs2_glock_nq(ghs); 898 if (error) 899 goto fail; 900 901 error = create_ok(dip, name, mode); 902 if (error) 903 goto fail_gunlock; 904 905 error = pick_formal_ino(sdp, &inum.no_formal_ino); 906 if (error) 907 goto fail_gunlock; 908 909 error = alloc_dinode(dip, &inum.no_addr, &generation); 910 if (error) 911 goto fail_gunlock; 912 913 error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops, 914 LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1); 915 if (error) 916 goto fail_gunlock; 917 918 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev); 919 if (error) 920 goto fail_gunlock2; 921 922 inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), 923 inum.no_addr, 924 inum.no_formal_ino); 925 if (IS_ERR(inode)) 926 goto fail_gunlock2; 927 928 error = gfs2_inode_refresh(GFS2_I(inode)); 929 if (error) 930 goto fail_gunlock2; 931 932 error = gfs2_acl_create(dip, GFS2_I(inode)); 933 if (error) 934 goto fail_gunlock2; 935 936 error = gfs2_security_init(dip, GFS2_I(inode)); 937 if (error) 938 goto fail_gunlock2; 939 940 error = link_dinode(dip, name, GFS2_I(inode)); 941 if (error) 942 goto fail_gunlock2; 943 944 if (!inode) 945 return ERR_PTR(-ENOMEM); 946 return inode; 947 948 fail_gunlock2: 949 gfs2_glock_dq_uninit(ghs + 1); 950 if (inode) 951 iput(inode); 952 fail_gunlock: 953 gfs2_glock_dq(ghs); 954 fail: 955 return ERR_PTR(error); 956 } 957 958 /** 959 * gfs2_rmdiri - Remove a directory 960 * @dip: The parent directory of the directory to be removed 961 * @name: The name of the directory to be removed 962 * @ip: The GFS2 inode of the directory to be removed 963 * 964 * Assumes Glocks on dip and ip are held 965 * 966 * Returns: errno 967 */ 968 969 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 970 struct gfs2_inode *ip) 971 { 972 struct qstr dotname; 973 int error; 974 975 if (ip->i_di.di_entries != 2) { 976 if (gfs2_consist_inode(ip)) 977 gfs2_dinode_print(ip); 978 return -EIO; 979 } 980 981 error = gfs2_dir_del(dip, name); 982 if (error) 983 return error; 984 985 error = gfs2_change_nlink(dip, -1); 986 if (error) 987 return error; 988 989 gfs2_str2qstr(&dotname, "."); 990 error = gfs2_dir_del(ip, &dotname); 991 if (error) 992 return error; 993 994 gfs2_str2qstr(&dotname, ".."); 995 error = gfs2_dir_del(ip, &dotname); 996 if (error) 997 return error; 998 999 /* It looks odd, but it really should be done twice */ 1000 error = gfs2_change_nlink(ip, -1); 1001 if (error) 1002 return error; 1003 1004 error = gfs2_change_nlink(ip, -1); 1005 if (error) 1006 return error; 1007 1008 return error; 1009 } 1010 1011 /* 1012 * gfs2_unlink_ok - check to see that a inode is still in a directory 1013 * @dip: the directory 1014 * @name: the name of the file 1015 * @ip: the inode 1016 * 1017 * Assumes that the lock on (at least) @dip is held. 1018 * 1019 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 1020 */ 1021 1022 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 1023 const struct gfs2_inode *ip) 1024 { 1025 int error; 1026 1027 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 1028 return -EPERM; 1029 1030 if ((dip->i_inode.i_mode & S_ISVTX) && 1031 dip->i_inode.i_uid != current->fsuid && 1032 ip->i_inode.i_uid != current->fsuid && !capable(CAP_FOWNER)) 1033 return -EPERM; 1034 1035 if (IS_APPEND(&dip->i_inode)) 1036 return -EPERM; 1037 1038 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 1039 if (error) 1040 return error; 1041 1042 error = gfs2_dir_check(&dip->i_inode, name, ip); 1043 if (error) 1044 return error; 1045 1046 return 0; 1047 } 1048 1049 /* 1050 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1051 * @this: move this 1052 * @to: to here 1053 * 1054 * Follow @to back to the root and make sure we don't encounter @this 1055 * Assumes we already hold the rename lock. 1056 * 1057 * Returns: errno 1058 */ 1059 1060 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1061 { 1062 struct inode *dir = &to->i_inode; 1063 struct super_block *sb = dir->i_sb; 1064 struct inode *tmp; 1065 struct qstr dotdot; 1066 int error = 0; 1067 1068 gfs2_str2qstr(&dotdot, ".."); 1069 1070 igrab(dir); 1071 1072 for (;;) { 1073 if (dir == &this->i_inode) { 1074 error = -EINVAL; 1075 break; 1076 } 1077 if (dir == sb->s_root->d_inode) { 1078 error = 0; 1079 break; 1080 } 1081 1082 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1083 if (IS_ERR(tmp)) { 1084 error = PTR_ERR(tmp); 1085 break; 1086 } 1087 1088 iput(dir); 1089 dir = tmp; 1090 } 1091 1092 iput(dir); 1093 1094 return error; 1095 } 1096 1097 /** 1098 * gfs2_readlinki - return the contents of a symlink 1099 * @ip: the symlink's inode 1100 * @buf: a pointer to the buffer to be filled 1101 * @len: a pointer to the length of @buf 1102 * 1103 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1104 * to be freed by the caller. 1105 * 1106 * Returns: errno 1107 */ 1108 1109 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1110 { 1111 struct gfs2_holder i_gh; 1112 struct buffer_head *dibh; 1113 unsigned int x; 1114 int error; 1115 1116 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1117 error = gfs2_glock_nq_atime(&i_gh); 1118 if (error) { 1119 gfs2_holder_uninit(&i_gh); 1120 return error; 1121 } 1122 1123 if (!ip->i_di.di_size) { 1124 gfs2_consist_inode(ip); 1125 error = -EIO; 1126 goto out; 1127 } 1128 1129 error = gfs2_meta_inode_buffer(ip, &dibh); 1130 if (error) 1131 goto out; 1132 1133 x = ip->i_di.di_size + 1; 1134 if (x > *len) { 1135 *buf = kmalloc(x, GFP_KERNEL); 1136 if (!*buf) { 1137 error = -ENOMEM; 1138 goto out_brelse; 1139 } 1140 } 1141 1142 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1143 *len = x; 1144 1145 out_brelse: 1146 brelse(dibh); 1147 out: 1148 gfs2_glock_dq_uninit(&i_gh); 1149 return error; 1150 } 1151 1152 /** 1153 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1154 * conditionally update the inode's atime 1155 * @gh: the holder to acquire 1156 * 1157 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1158 * Update if the difference between the current time and the inode's current 1159 * atime is greater than an interval specified at mount. 1160 * 1161 * Returns: errno 1162 */ 1163 1164 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1165 { 1166 struct gfs2_glock *gl = gh->gh_gl; 1167 struct gfs2_sbd *sdp = gl->gl_sbd; 1168 struct gfs2_inode *ip = gl->gl_object; 1169 s64 quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1170 unsigned int state; 1171 int flags; 1172 int error; 1173 struct timespec tv = CURRENT_TIME; 1174 1175 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1176 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1177 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1178 return -EINVAL; 1179 1180 state = gh->gh_state; 1181 flags = gh->gh_flags; 1182 1183 error = gfs2_glock_nq(gh); 1184 if (error) 1185 return error; 1186 1187 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1188 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1189 return 0; 1190 1191 if (tv.tv_sec - ip->i_inode.i_atime.tv_sec >= quantum) { 1192 gfs2_glock_dq(gh); 1193 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1194 gh); 1195 error = gfs2_glock_nq(gh); 1196 if (error) 1197 return error; 1198 1199 /* Verify that atime hasn't been updated while we were 1200 trying to get exclusive lock. */ 1201 1202 tv = CURRENT_TIME; 1203 if (tv.tv_sec - ip->i_inode.i_atime.tv_sec >= quantum) { 1204 struct buffer_head *dibh; 1205 struct gfs2_dinode *di; 1206 1207 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1208 if (error == -EROFS) 1209 return 0; 1210 if (error) 1211 goto fail; 1212 1213 error = gfs2_meta_inode_buffer(ip, &dibh); 1214 if (error) 1215 goto fail_end_trans; 1216 1217 ip->i_inode.i_atime = tv; 1218 1219 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1220 di = (struct gfs2_dinode *)dibh->b_data; 1221 di->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec); 1222 di->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec); 1223 brelse(dibh); 1224 1225 gfs2_trans_end(sdp); 1226 } 1227 1228 /* If someone else has asked for the glock, 1229 unlock and let them have it. Then reacquire 1230 in the original state. */ 1231 if (gfs2_glock_is_blocking(gl)) { 1232 gfs2_glock_dq(gh); 1233 gfs2_holder_reinit(state, flags, gh); 1234 return gfs2_glock_nq(gh); 1235 } 1236 } 1237 1238 return 0; 1239 1240 fail_end_trans: 1241 gfs2_trans_end(sdp); 1242 fail: 1243 gfs2_glock_dq(gh); 1244 return error; 1245 } 1246 1247 static int 1248 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1249 { 1250 struct buffer_head *dibh; 1251 int error; 1252 1253 error = gfs2_meta_inode_buffer(ip, &dibh); 1254 if (!error) { 1255 error = inode_setattr(&ip->i_inode, attr); 1256 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1257 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1258 gfs2_dinode_out(ip, dibh->b_data); 1259 brelse(dibh); 1260 } 1261 return error; 1262 } 1263 1264 /** 1265 * gfs2_setattr_simple - 1266 * @ip: 1267 * @attr: 1268 * 1269 * Called with a reference on the vnode. 1270 * 1271 * Returns: errno 1272 */ 1273 1274 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1275 { 1276 int error; 1277 1278 if (current->journal_info) 1279 return __gfs2_setattr_simple(ip, attr); 1280 1281 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1282 if (error) 1283 return error; 1284 1285 error = __gfs2_setattr_simple(ip, attr); 1286 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1287 return error; 1288 } 1289 1290 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf) 1291 { 1292 const struct gfs2_dinode_host *di = &ip->i_di; 1293 struct gfs2_dinode *str = buf; 1294 1295 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 1296 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI); 1297 str->di_header.__pad0 = 0; 1298 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI); 1299 str->di_header.__pad1 = 0; 1300 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr); 1301 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino); 1302 str->di_mode = cpu_to_be32(ip->i_inode.i_mode); 1303 str->di_uid = cpu_to_be32(ip->i_inode.i_uid); 1304 str->di_gid = cpu_to_be32(ip->i_inode.i_gid); 1305 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink); 1306 str->di_size = cpu_to_be64(di->di_size); 1307 str->di_blocks = cpu_to_be64(di->di_blocks); 1308 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec); 1309 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec); 1310 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec); 1311 1312 str->di_goal_meta = cpu_to_be64(di->di_goal_meta); 1313 str->di_goal_data = cpu_to_be64(di->di_goal_data); 1314 str->di_generation = cpu_to_be64(di->di_generation); 1315 1316 str->di_flags = cpu_to_be32(di->di_flags); 1317 str->di_height = cpu_to_be16(di->di_height); 1318 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) && 1319 !(ip->i_di.di_flags & GFS2_DIF_EXHASH) ? 1320 GFS2_FORMAT_DE : 0); 1321 str->di_depth = cpu_to_be16(di->di_depth); 1322 str->di_entries = cpu_to_be32(di->di_entries); 1323 1324 str->di_eattr = cpu_to_be64(di->di_eattr); 1325 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec); 1326 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec); 1327 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec); 1328 } 1329 1330 void gfs2_dinode_print(const struct gfs2_inode *ip) 1331 { 1332 const struct gfs2_dinode_host *di = &ip->i_di; 1333 1334 printk(KERN_INFO " no_formal_ino = %llu\n", 1335 (unsigned long long)ip->i_no_formal_ino); 1336 printk(KERN_INFO " no_addr = %llu\n", 1337 (unsigned long long)ip->i_no_addr); 1338 printk(KERN_INFO " di_size = %llu\n", (unsigned long long)di->di_size); 1339 printk(KERN_INFO " di_blocks = %llu\n", 1340 (unsigned long long)di->di_blocks); 1341 printk(KERN_INFO " di_goal_meta = %llu\n", 1342 (unsigned long long)di->di_goal_meta); 1343 printk(KERN_INFO " di_goal_data = %llu\n", 1344 (unsigned long long)di->di_goal_data); 1345 printk(KERN_INFO " di_flags = 0x%.8X\n", di->di_flags); 1346 printk(KERN_INFO " di_height = %u\n", di->di_height); 1347 printk(KERN_INFO " di_depth = %u\n", di->di_depth); 1348 printk(KERN_INFO " di_entries = %u\n", di->di_entries); 1349 printk(KERN_INFO " di_eattr = %llu\n", 1350 (unsigned long long)di->di_eattr); 1351 } 1352 1353