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 /** 42 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode 43 * @ip: The GFS2 inode (with embedded disk inode data) 44 * @inode: The Linux VFS inode 45 * 46 */ 47 48 void gfs2_inode_attr_in(struct gfs2_inode *ip) 49 { 50 struct inode *inode = &ip->i_inode; 51 struct gfs2_dinode_host *di = &ip->i_di; 52 53 inode->i_ino = ip->i_num.no_addr; 54 i_size_write(inode, di->di_size); 55 inode->i_atime.tv_sec = di->di_atime; 56 inode->i_mtime.tv_sec = di->di_mtime; 57 inode->i_ctime.tv_sec = di->di_ctime; 58 inode->i_atime.tv_nsec = 0; 59 inode->i_mtime.tv_nsec = 0; 60 inode->i_ctime.tv_nsec = 0; 61 inode->i_blocks = di->di_blocks << 62 (GFS2_SB(inode)->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT); 63 64 if (di->di_flags & GFS2_DIF_IMMUTABLE) 65 inode->i_flags |= S_IMMUTABLE; 66 else 67 inode->i_flags &= ~S_IMMUTABLE; 68 69 if (di->di_flags & GFS2_DIF_APPENDONLY) 70 inode->i_flags |= S_APPEND; 71 else 72 inode->i_flags &= ~S_APPEND; 73 } 74 75 /** 76 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode 77 * @ip: The GFS2 inode 78 * 79 * Only copy out the attributes that we want the VFS layer 80 * to be able to modify. 81 */ 82 83 void gfs2_inode_attr_out(struct gfs2_inode *ip) 84 { 85 struct inode *inode = &ip->i_inode; 86 struct gfs2_dinode_host *di = &ip->i_di; 87 di->di_atime = inode->i_atime.tv_sec; 88 di->di_mtime = inode->i_mtime.tv_sec; 89 di->di_ctime = inode->i_ctime.tv_sec; 90 } 91 92 static int iget_test(struct inode *inode, void *opaque) 93 { 94 struct gfs2_inode *ip = GFS2_I(inode); 95 struct gfs2_inum_host *inum = opaque; 96 97 if (ip && ip->i_num.no_addr == inum->no_addr) 98 return 1; 99 100 return 0; 101 } 102 103 static int iget_set(struct inode *inode, void *opaque) 104 { 105 struct gfs2_inode *ip = GFS2_I(inode); 106 struct gfs2_inum_host *inum = opaque; 107 108 ip->i_num = *inum; 109 return 0; 110 } 111 112 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum_host *inum) 113 { 114 return ilookup5(sb, (unsigned long)inum->no_formal_ino, 115 iget_test, inum); 116 } 117 118 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum_host *inum) 119 { 120 return iget5_locked(sb, (unsigned long)inum->no_formal_ino, 121 iget_test, iget_set, inum); 122 } 123 124 /** 125 * gfs2_inode_lookup - Lookup an inode 126 * @sb: The super block 127 * @inum: The inode number 128 * @type: The type of the inode 129 * 130 * Returns: A VFS inode, or an error 131 */ 132 133 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum_host *inum, unsigned int type) 134 { 135 struct inode *inode = gfs2_iget(sb, inum); 136 struct gfs2_inode *ip = GFS2_I(inode); 137 struct gfs2_glock *io_gl; 138 int error; 139 140 if (!inode) 141 return ERR_PTR(-ENOBUFS); 142 143 if (inode->i_state & I_NEW) { 144 struct gfs2_sbd *sdp = GFS2_SB(inode); 145 umode_t mode = DT2IF(type); 146 inode->i_private = ip; 147 inode->i_mode = mode; 148 149 if (S_ISREG(mode)) { 150 inode->i_op = &gfs2_file_iops; 151 inode->i_fop = &gfs2_file_fops; 152 inode->i_mapping->a_ops = &gfs2_file_aops; 153 } else if (S_ISDIR(mode)) { 154 inode->i_op = &gfs2_dir_iops; 155 inode->i_fop = &gfs2_dir_fops; 156 } else if (S_ISLNK(mode)) { 157 inode->i_op = &gfs2_symlink_iops; 158 } else { 159 inode->i_op = &gfs2_dev_iops; 160 } 161 162 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl); 163 if (unlikely(error)) 164 goto fail; 165 ip->i_gl->gl_object = ip; 166 167 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl); 168 if (unlikely(error)) 169 goto fail_put; 170 171 ip->i_vn = ip->i_gl->gl_vn - 1; 172 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh); 173 if (unlikely(error)) 174 goto fail_iopen; 175 176 gfs2_glock_put(io_gl); 177 unlock_new_inode(inode); 178 } 179 180 return inode; 181 fail_iopen: 182 gfs2_glock_put(io_gl); 183 fail_put: 184 ip->i_gl->gl_object = NULL; 185 gfs2_glock_put(ip->i_gl); 186 fail: 187 iput(inode); 188 return ERR_PTR(error); 189 } 190 191 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf) 192 { 193 struct gfs2_dinode_host *di = &ip->i_di; 194 const struct gfs2_dinode *str = buf; 195 196 if (ip->i_num.no_addr != be64_to_cpu(str->di_num.no_addr)) { 197 if (gfs2_consist_inode(ip)) 198 gfs2_dinode_print(ip); 199 return -EIO; 200 } 201 if (ip->i_num.no_formal_ino != be64_to_cpu(str->di_num.no_formal_ino)) 202 return -ESTALE; 203 204 ip->i_inode.i_mode = be32_to_cpu(str->di_mode); 205 ip->i_inode.i_rdev = 0; 206 switch (ip->i_inode.i_mode & S_IFMT) { 207 case S_IFBLK: 208 case S_IFCHR: 209 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major), 210 be32_to_cpu(str->di_minor)); 211 break; 212 }; 213 214 ip->i_inode.i_uid = be32_to_cpu(str->di_uid); 215 ip->i_inode.i_gid = be32_to_cpu(str->di_gid); 216 /* 217 * We will need to review setting the nlink count here in the 218 * light of the forthcoming ro bind mount work. This is a reminder 219 * to do that. 220 */ 221 ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink); 222 di->di_size = be64_to_cpu(str->di_size); 223 di->di_blocks = be64_to_cpu(str->di_blocks); 224 di->di_atime = be64_to_cpu(str->di_atime); 225 di->di_mtime = be64_to_cpu(str->di_mtime); 226 di->di_ctime = be64_to_cpu(str->di_ctime); 227 228 di->di_goal_meta = be64_to_cpu(str->di_goal_meta); 229 di->di_goal_data = be64_to_cpu(str->di_goal_data); 230 di->di_generation = be64_to_cpu(str->di_generation); 231 232 di->di_flags = be32_to_cpu(str->di_flags); 233 di->di_payload_format = be32_to_cpu(str->di_payload_format); 234 di->di_height = be16_to_cpu(str->di_height); 235 236 di->di_depth = be16_to_cpu(str->di_depth); 237 di->di_entries = be32_to_cpu(str->di_entries); 238 239 di->di_eattr = be64_to_cpu(str->di_eattr); 240 return 0; 241 } 242 243 /** 244 * gfs2_inode_refresh - Refresh the incore copy of the dinode 245 * @ip: The GFS2 inode 246 * 247 * Returns: errno 248 */ 249 250 int gfs2_inode_refresh(struct gfs2_inode *ip) 251 { 252 struct buffer_head *dibh; 253 int error; 254 255 error = gfs2_meta_inode_buffer(ip, &dibh); 256 if (error) 257 return error; 258 259 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) { 260 brelse(dibh); 261 return -EIO; 262 } 263 264 error = gfs2_dinode_in(ip, dibh->b_data); 265 brelse(dibh); 266 ip->i_vn = ip->i_gl->gl_vn; 267 268 return error; 269 } 270 271 int gfs2_dinode_dealloc(struct gfs2_inode *ip) 272 { 273 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 274 struct gfs2_alloc *al; 275 struct gfs2_rgrpd *rgd; 276 int error; 277 278 if (ip->i_di.di_blocks != 1) { 279 if (gfs2_consist_inode(ip)) 280 gfs2_dinode_print(ip); 281 return -EIO; 282 } 283 284 al = gfs2_alloc_get(ip); 285 286 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 287 if (error) 288 goto out; 289 290 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 291 if (error) 292 goto out_qs; 293 294 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 295 if (!rgd) { 296 gfs2_consist_inode(ip); 297 error = -EIO; 298 goto out_rindex_relse; 299 } 300 301 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 302 &al->al_rgd_gh); 303 if (error) 304 goto out_rindex_relse; 305 306 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1); 307 if (error) 308 goto out_rg_gunlock; 309 310 gfs2_trans_add_gl(ip->i_gl); 311 312 gfs2_free_di(rgd, ip); 313 314 gfs2_trans_end(sdp); 315 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 316 317 out_rg_gunlock: 318 gfs2_glock_dq_uninit(&al->al_rgd_gh); 319 out_rindex_relse: 320 gfs2_glock_dq_uninit(&al->al_ri_gh); 321 out_qs: 322 gfs2_quota_unhold(ip); 323 out: 324 gfs2_alloc_put(ip); 325 return error; 326 } 327 328 /** 329 * gfs2_change_nlink - Change nlink count on inode 330 * @ip: The GFS2 inode 331 * @diff: The change in the nlink count required 332 * 333 * Returns: errno 334 */ 335 336 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 337 { 338 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info; 339 struct buffer_head *dibh; 340 u32 nlink; 341 int error; 342 343 BUG_ON(diff != 1 && diff != -1); 344 nlink = ip->i_inode.i_nlink + diff; 345 346 /* If we are reducing the nlink count, but the new value ends up being 347 bigger than the old one, we must have underflowed. */ 348 if (diff < 0 && nlink > ip->i_inode.i_nlink) { 349 if (gfs2_consist_inode(ip)) 350 gfs2_dinode_print(ip); 351 return -EIO; 352 } 353 354 error = gfs2_meta_inode_buffer(ip, &dibh); 355 if (error) 356 return error; 357 358 if (diff > 0) 359 inc_nlink(&ip->i_inode); 360 else 361 drop_nlink(&ip->i_inode); 362 363 ip->i_di.di_ctime = get_seconds(); 364 365 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 366 gfs2_dinode_out(ip, dibh->b_data); 367 brelse(dibh); 368 mark_inode_dirty(&ip->i_inode); 369 370 if (ip->i_inode.i_nlink == 0) { 371 struct gfs2_rgrpd *rgd; 372 struct gfs2_holder ri_gh, rg_gh; 373 374 error = gfs2_rindex_hold(sdp, &ri_gh); 375 if (error) 376 goto out; 377 error = -EIO; 378 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 379 if (!rgd) 380 goto out_norgrp; 381 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh); 382 if (error) 383 goto out_norgrp; 384 385 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 386 gfs2_glock_dq_uninit(&rg_gh); 387 out_norgrp: 388 gfs2_glock_dq_uninit(&ri_gh); 389 } 390 out: 391 return error; 392 } 393 394 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 395 { 396 struct qstr qstr; 397 gfs2_str2qstr(&qstr, name); 398 return gfs2_lookupi(dip, &qstr, 1, NULL); 399 } 400 401 402 /** 403 * gfs2_lookupi - Look up a filename in a directory and return its inode 404 * @d_gh: An initialized holder for the directory glock 405 * @name: The name of the inode to look for 406 * @is_root: If 1, ignore the caller's permissions 407 * @i_gh: An uninitialized holder for the new inode glock 408 * 409 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 410 * @is_root is true. 411 * 412 * Returns: errno 413 */ 414 415 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 416 int is_root, struct nameidata *nd) 417 { 418 struct super_block *sb = dir->i_sb; 419 struct gfs2_inode *dip = GFS2_I(dir); 420 struct gfs2_holder d_gh; 421 struct gfs2_inum_host inum; 422 unsigned int type; 423 int error = 0; 424 struct inode *inode = NULL; 425 426 if (!name->len || name->len > GFS2_FNAMESIZE) 427 return ERR_PTR(-ENAMETOOLONG); 428 429 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 430 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 431 dir == sb->s_root->d_inode)) { 432 igrab(dir); 433 return dir; 434 } 435 436 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 437 if (error) 438 return ERR_PTR(error); 439 440 if (!is_root) { 441 error = permission(dir, MAY_EXEC, NULL); 442 if (error) 443 goto out; 444 } 445 446 error = gfs2_dir_search(dir, name, &inum, &type); 447 if (error) 448 goto out; 449 450 inode = gfs2_inode_lookup(sb, &inum, type); 451 452 out: 453 gfs2_glock_dq_uninit(&d_gh); 454 if (error == -ENOENT) 455 return NULL; 456 return inode; 457 } 458 459 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino) 460 { 461 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 462 struct buffer_head *bh; 463 struct gfs2_inum_range_host ir; 464 int error; 465 466 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 467 if (error) 468 return error; 469 mutex_lock(&sdp->sd_inum_mutex); 470 471 error = gfs2_meta_inode_buffer(ip, &bh); 472 if (error) { 473 mutex_unlock(&sdp->sd_inum_mutex); 474 gfs2_trans_end(sdp); 475 return error; 476 } 477 478 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 479 480 if (ir.ir_length) { 481 *formal_ino = ir.ir_start++; 482 ir.ir_length--; 483 gfs2_trans_add_bh(ip->i_gl, bh, 1); 484 gfs2_inum_range_out(&ir, 485 bh->b_data + sizeof(struct gfs2_dinode)); 486 brelse(bh); 487 mutex_unlock(&sdp->sd_inum_mutex); 488 gfs2_trans_end(sdp); 489 return 0; 490 } 491 492 brelse(bh); 493 494 mutex_unlock(&sdp->sd_inum_mutex); 495 gfs2_trans_end(sdp); 496 497 return 1; 498 } 499 500 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino) 501 { 502 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 503 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 504 struct gfs2_holder gh; 505 struct buffer_head *bh; 506 struct gfs2_inum_range_host ir; 507 int error; 508 509 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 510 if (error) 511 return error; 512 513 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 514 if (error) 515 goto out; 516 mutex_lock(&sdp->sd_inum_mutex); 517 518 error = gfs2_meta_inode_buffer(ip, &bh); 519 if (error) 520 goto out_end_trans; 521 522 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 523 524 if (!ir.ir_length) { 525 struct buffer_head *m_bh; 526 u64 x, y; 527 __be64 z; 528 529 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 530 if (error) 531 goto out_brelse; 532 533 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 534 x = y = be64_to_cpu(z); 535 ir.ir_start = x; 536 ir.ir_length = GFS2_INUM_QUANTUM; 537 x += GFS2_INUM_QUANTUM; 538 if (x < y) 539 gfs2_consist_inode(m_ip); 540 z = cpu_to_be64(x); 541 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 542 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z; 543 544 brelse(m_bh); 545 } 546 547 *formal_ino = ir.ir_start++; 548 ir.ir_length--; 549 550 gfs2_trans_add_bh(ip->i_gl, bh, 1); 551 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 552 553 out_brelse: 554 brelse(bh); 555 out_end_trans: 556 mutex_unlock(&sdp->sd_inum_mutex); 557 gfs2_trans_end(sdp); 558 out: 559 gfs2_glock_dq_uninit(&gh); 560 return error; 561 } 562 563 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum) 564 { 565 int error; 566 567 error = pick_formal_ino_1(sdp, inum); 568 if (error <= 0) 569 return error; 570 571 error = pick_formal_ino_2(sdp, inum); 572 573 return error; 574 } 575 576 /** 577 * create_ok - OK to create a new on-disk inode here? 578 * @dip: Directory in which dinode is to be created 579 * @name: Name of new dinode 580 * @mode: 581 * 582 * Returns: errno 583 */ 584 585 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 586 unsigned int mode) 587 { 588 int error; 589 590 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 591 if (error) 592 return error; 593 594 /* Don't create entries in an unlinked directory */ 595 if (!dip->i_inode.i_nlink) 596 return -EPERM; 597 598 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL); 599 switch (error) { 600 case -ENOENT: 601 error = 0; 602 break; 603 case 0: 604 return -EEXIST; 605 default: 606 return error; 607 } 608 609 if (dip->i_di.di_entries == (u32)-1) 610 return -EFBIG; 611 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1) 612 return -EMLINK; 613 614 return 0; 615 } 616 617 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 618 unsigned int *uid, unsigned int *gid) 619 { 620 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 621 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) { 622 if (S_ISDIR(*mode)) 623 *mode |= S_ISUID; 624 else if (dip->i_inode.i_uid != current->fsuid) 625 *mode &= ~07111; 626 *uid = dip->i_inode.i_uid; 627 } else 628 *uid = current->fsuid; 629 630 if (dip->i_inode.i_mode & S_ISGID) { 631 if (S_ISDIR(*mode)) 632 *mode |= S_ISGID; 633 *gid = dip->i_inode.i_gid; 634 } else 635 *gid = current->fsgid; 636 } 637 638 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum, 639 u64 *generation) 640 { 641 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 642 int error; 643 644 gfs2_alloc_get(dip); 645 646 dip->i_alloc.al_requested = RES_DINODE; 647 error = gfs2_inplace_reserve(dip); 648 if (error) 649 goto out; 650 651 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 652 if (error) 653 goto out_ipreserv; 654 655 inum->no_addr = gfs2_alloc_di(dip, generation); 656 657 gfs2_trans_end(sdp); 658 659 out_ipreserv: 660 gfs2_inplace_release(dip); 661 out: 662 gfs2_alloc_put(dip); 663 return error; 664 } 665 666 /** 667 * init_dinode - Fill in a new dinode structure 668 * @dip: the directory this inode is being created in 669 * @gl: The glock covering the new inode 670 * @inum: the inode number 671 * @mode: the file permissions 672 * @uid: 673 * @gid: 674 * 675 */ 676 677 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 678 const struct gfs2_inum_host *inum, unsigned int mode, 679 unsigned int uid, unsigned int gid, 680 const u64 *generation, dev_t dev) 681 { 682 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 683 struct gfs2_dinode *di; 684 struct buffer_head *dibh; 685 686 dibh = gfs2_meta_new(gl, inum->no_addr); 687 gfs2_trans_add_bh(gl, dibh, 1); 688 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 689 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 690 di = (struct gfs2_dinode *)dibh->b_data; 691 692 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 693 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 694 di->di_mode = cpu_to_be32(mode); 695 di->di_uid = cpu_to_be32(uid); 696 di->di_gid = cpu_to_be32(gid); 697 di->di_nlink = cpu_to_be32(0); 698 di->di_size = cpu_to_be64(0); 699 di->di_blocks = cpu_to_be64(1); 700 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 701 di->di_major = cpu_to_be32(MAJOR(dev)); 702 di->di_minor = cpu_to_be32(MINOR(dev)); 703 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 704 di->di_generation = cpu_to_be64(*generation); 705 di->di_flags = cpu_to_be32(0); 706 707 if (S_ISREG(mode)) { 708 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 709 gfs2_tune_get(sdp, gt_new_files_jdata)) 710 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 711 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 712 gfs2_tune_get(sdp, gt_new_files_directio)) 713 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 714 } else if (S_ISDIR(mode)) { 715 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 716 GFS2_DIF_INHERIT_DIRECTIO); 717 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 718 GFS2_DIF_INHERIT_JDATA); 719 } 720 721 di->__pad1 = 0; 722 di->di_payload_format = cpu_to_be32(0); 723 di->di_height = cpu_to_be32(0); 724 di->__pad2 = 0; 725 di->__pad3 = 0; 726 di->di_depth = cpu_to_be16(0); 727 di->di_entries = cpu_to_be32(0); 728 memset(&di->__pad4, 0, sizeof(di->__pad4)); 729 di->di_eattr = cpu_to_be64(0); 730 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 731 732 brelse(dibh); 733 } 734 735 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 736 unsigned int mode, const struct gfs2_inum_host *inum, 737 const u64 *generation, dev_t dev) 738 { 739 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 740 unsigned int uid, gid; 741 int error; 742 743 munge_mode_uid_gid(dip, &mode, &uid, &gid); 744 gfs2_alloc_get(dip); 745 746 error = gfs2_quota_lock(dip, uid, gid); 747 if (error) 748 goto out; 749 750 error = gfs2_quota_check(dip, uid, gid); 751 if (error) 752 goto out_quota; 753 754 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0); 755 if (error) 756 goto out_quota; 757 758 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev); 759 gfs2_quota_change(dip, +1, uid, gid); 760 gfs2_trans_end(sdp); 761 762 out_quota: 763 gfs2_quota_unlock(dip); 764 out: 765 gfs2_alloc_put(dip); 766 return error; 767 } 768 769 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name, 770 struct gfs2_inode *ip) 771 { 772 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 773 struct gfs2_alloc *al; 774 int alloc_required; 775 struct buffer_head *dibh; 776 int error; 777 778 al = gfs2_alloc_get(dip); 779 780 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 781 if (error) 782 goto fail; 783 784 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name); 785 if (alloc_required < 0) 786 goto fail; 787 if (alloc_required) { 788 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid); 789 if (error) 790 goto fail_quota_locks; 791 792 al->al_requested = sdp->sd_max_dirres; 793 794 error = gfs2_inplace_reserve(dip); 795 if (error) 796 goto fail_quota_locks; 797 798 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 799 al->al_rgd->rd_ri.ri_length + 800 2 * RES_DINODE + 801 RES_STATFS + RES_QUOTA, 0); 802 if (error) 803 goto fail_ipreserv; 804 } else { 805 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 806 if (error) 807 goto fail_quota_locks; 808 } 809 810 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_inode.i_mode)); 811 if (error) 812 goto fail_end_trans; 813 814 error = gfs2_meta_inode_buffer(ip, &dibh); 815 if (error) 816 goto fail_end_trans; 817 ip->i_inode.i_nlink = 1; 818 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 819 gfs2_dinode_out(ip, dibh->b_data); 820 brelse(dibh); 821 return 0; 822 823 fail_end_trans: 824 gfs2_trans_end(sdp); 825 826 fail_ipreserv: 827 if (dip->i_alloc.al_rgd) 828 gfs2_inplace_release(dip); 829 830 fail_quota_locks: 831 gfs2_quota_unlock(dip); 832 833 fail: 834 gfs2_alloc_put(dip); 835 return error; 836 } 837 838 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip) 839 { 840 int err; 841 size_t len; 842 void *value; 843 char *name; 844 struct gfs2_ea_request er; 845 846 err = security_inode_init_security(&ip->i_inode, &dip->i_inode, 847 &name, &value, &len); 848 849 if (err) { 850 if (err == -EOPNOTSUPP) 851 return 0; 852 return err; 853 } 854 855 memset(&er, 0, sizeof(struct gfs2_ea_request)); 856 857 er.er_type = GFS2_EATYPE_SECURITY; 858 er.er_name = name; 859 er.er_data = value; 860 er.er_name_len = strlen(name); 861 er.er_data_len = len; 862 863 err = gfs2_ea_set_i(ip, &er); 864 865 kfree(value); 866 kfree(name); 867 868 return err; 869 } 870 871 /** 872 * gfs2_createi - Create a new inode 873 * @ghs: An array of two holders 874 * @name: The name of the new file 875 * @mode: the permissions on the new inode 876 * 877 * @ghs[0] is an initialized holder for the directory 878 * @ghs[1] is the holder for the inode lock 879 * 880 * If the return value is not NULL, the glocks on both the directory and the new 881 * file are held. A transaction has been started and an inplace reservation 882 * is held, as well. 883 * 884 * Returns: An inode 885 */ 886 887 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 888 unsigned int mode, dev_t dev) 889 { 890 struct inode *inode; 891 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 892 struct inode *dir = &dip->i_inode; 893 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 894 struct gfs2_inum_host inum; 895 int error; 896 u64 generation; 897 898 if (!name->len || name->len > GFS2_FNAMESIZE) 899 return ERR_PTR(-ENAMETOOLONG); 900 901 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 902 error = gfs2_glock_nq(ghs); 903 if (error) 904 goto fail; 905 906 error = create_ok(dip, name, mode); 907 if (error) 908 goto fail_gunlock; 909 910 error = pick_formal_ino(sdp, &inum.no_formal_ino); 911 if (error) 912 goto fail_gunlock; 913 914 error = alloc_dinode(dip, &inum, &generation); 915 if (error) 916 goto fail_gunlock; 917 918 if (inum.no_addr < dip->i_num.no_addr) { 919 gfs2_glock_dq(ghs); 920 921 error = gfs2_glock_nq_num(sdp, inum.no_addr, 922 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 923 GL_SKIP, ghs + 1); 924 if (error) { 925 return ERR_PTR(error); 926 } 927 928 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 929 error = gfs2_glock_nq(ghs); 930 if (error) { 931 gfs2_glock_dq_uninit(ghs + 1); 932 return ERR_PTR(error); 933 } 934 935 error = create_ok(dip, name, mode); 936 if (error) 937 goto fail_gunlock2; 938 } else { 939 error = gfs2_glock_nq_num(sdp, inum.no_addr, 940 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 941 GL_SKIP, ghs + 1); 942 if (error) 943 goto fail_gunlock; 944 } 945 946 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev); 947 if (error) 948 goto fail_gunlock2; 949 950 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode)); 951 if (IS_ERR(inode)) 952 goto fail_gunlock2; 953 954 error = gfs2_inode_refresh(GFS2_I(inode)); 955 if (error) 956 goto fail_iput; 957 958 error = gfs2_acl_create(dip, GFS2_I(inode)); 959 if (error) 960 goto fail_iput; 961 962 error = gfs2_security_init(dip, GFS2_I(inode)); 963 if (error) 964 goto fail_iput; 965 966 error = link_dinode(dip, name, GFS2_I(inode)); 967 if (error) 968 goto fail_iput; 969 970 if (!inode) 971 return ERR_PTR(-ENOMEM); 972 return inode; 973 974 fail_iput: 975 iput(inode); 976 fail_gunlock2: 977 gfs2_glock_dq_uninit(ghs + 1); 978 fail_gunlock: 979 gfs2_glock_dq(ghs); 980 fail: 981 return ERR_PTR(error); 982 } 983 984 /** 985 * gfs2_rmdiri - Remove a directory 986 * @dip: The parent directory of the directory to be removed 987 * @name: The name of the directory to be removed 988 * @ip: The GFS2 inode of the directory to be removed 989 * 990 * Assumes Glocks on dip and ip are held 991 * 992 * Returns: errno 993 */ 994 995 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 996 struct gfs2_inode *ip) 997 { 998 struct qstr dotname; 999 int error; 1000 1001 if (ip->i_di.di_entries != 2) { 1002 if (gfs2_consist_inode(ip)) 1003 gfs2_dinode_print(ip); 1004 return -EIO; 1005 } 1006 1007 error = gfs2_dir_del(dip, name); 1008 if (error) 1009 return error; 1010 1011 error = gfs2_change_nlink(dip, -1); 1012 if (error) 1013 return error; 1014 1015 gfs2_str2qstr(&dotname, "."); 1016 error = gfs2_dir_del(ip, &dotname); 1017 if (error) 1018 return error; 1019 1020 gfs2_str2qstr(&dotname, ".."); 1021 error = gfs2_dir_del(ip, &dotname); 1022 if (error) 1023 return error; 1024 1025 /* It looks odd, but it really should be done twice */ 1026 error = gfs2_change_nlink(ip, -1); 1027 if (error) 1028 return error; 1029 1030 error = gfs2_change_nlink(ip, -1); 1031 if (error) 1032 return error; 1033 1034 return error; 1035 } 1036 1037 /* 1038 * gfs2_unlink_ok - check to see that a inode is still in a directory 1039 * @dip: the directory 1040 * @name: the name of the file 1041 * @ip: the inode 1042 * 1043 * Assumes that the lock on (at least) @dip is held. 1044 * 1045 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 1046 */ 1047 1048 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 1049 struct gfs2_inode *ip) 1050 { 1051 struct gfs2_inum_host inum; 1052 unsigned int type; 1053 int error; 1054 1055 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 1056 return -EPERM; 1057 1058 if ((dip->i_inode.i_mode & S_ISVTX) && 1059 dip->i_inode.i_uid != current->fsuid && 1060 ip->i_inode.i_uid != current->fsuid && !capable(CAP_FOWNER)) 1061 return -EPERM; 1062 1063 if (IS_APPEND(&dip->i_inode)) 1064 return -EPERM; 1065 1066 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 1067 if (error) 1068 return error; 1069 1070 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type); 1071 if (error) 1072 return error; 1073 1074 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1075 return -ENOENT; 1076 1077 if (IF2DT(ip->i_inode.i_mode) != type) { 1078 gfs2_consist_inode(dip); 1079 return -EIO; 1080 } 1081 1082 return 0; 1083 } 1084 1085 /* 1086 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1087 * @this: move this 1088 * @to: to here 1089 * 1090 * Follow @to back to the root and make sure we don't encounter @this 1091 * Assumes we already hold the rename lock. 1092 * 1093 * Returns: errno 1094 */ 1095 1096 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1097 { 1098 struct inode *dir = &to->i_inode; 1099 struct super_block *sb = dir->i_sb; 1100 struct inode *tmp; 1101 struct qstr dotdot; 1102 int error = 0; 1103 1104 gfs2_str2qstr(&dotdot, ".."); 1105 1106 igrab(dir); 1107 1108 for (;;) { 1109 if (dir == &this->i_inode) { 1110 error = -EINVAL; 1111 break; 1112 } 1113 if (dir == sb->s_root->d_inode) { 1114 error = 0; 1115 break; 1116 } 1117 1118 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1119 if (IS_ERR(tmp)) { 1120 error = PTR_ERR(tmp); 1121 break; 1122 } 1123 1124 iput(dir); 1125 dir = tmp; 1126 } 1127 1128 iput(dir); 1129 1130 return error; 1131 } 1132 1133 /** 1134 * gfs2_readlinki - return the contents of a symlink 1135 * @ip: the symlink's inode 1136 * @buf: a pointer to the buffer to be filled 1137 * @len: a pointer to the length of @buf 1138 * 1139 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1140 * to be freed by the caller. 1141 * 1142 * Returns: errno 1143 */ 1144 1145 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1146 { 1147 struct gfs2_holder i_gh; 1148 struct buffer_head *dibh; 1149 unsigned int x; 1150 int error; 1151 1152 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1153 error = gfs2_glock_nq_atime(&i_gh); 1154 if (error) { 1155 gfs2_holder_uninit(&i_gh); 1156 return error; 1157 } 1158 1159 if (!ip->i_di.di_size) { 1160 gfs2_consist_inode(ip); 1161 error = -EIO; 1162 goto out; 1163 } 1164 1165 error = gfs2_meta_inode_buffer(ip, &dibh); 1166 if (error) 1167 goto out; 1168 1169 x = ip->i_di.di_size + 1; 1170 if (x > *len) { 1171 *buf = kmalloc(x, GFP_KERNEL); 1172 if (!*buf) { 1173 error = -ENOMEM; 1174 goto out_brelse; 1175 } 1176 } 1177 1178 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1179 *len = x; 1180 1181 out_brelse: 1182 brelse(dibh); 1183 out: 1184 gfs2_glock_dq_uninit(&i_gh); 1185 return error; 1186 } 1187 1188 /** 1189 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1190 * conditionally update the inode's atime 1191 * @gh: the holder to acquire 1192 * 1193 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1194 * Update if the difference between the current time and the inode's current 1195 * atime is greater than an interval specified at mount. 1196 * 1197 * Returns: errno 1198 */ 1199 1200 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1201 { 1202 struct gfs2_glock *gl = gh->gh_gl; 1203 struct gfs2_sbd *sdp = gl->gl_sbd; 1204 struct gfs2_inode *ip = gl->gl_object; 1205 s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1206 unsigned int state; 1207 int flags; 1208 int error; 1209 1210 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1211 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1212 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1213 return -EINVAL; 1214 1215 state = gh->gh_state; 1216 flags = gh->gh_flags; 1217 1218 error = gfs2_glock_nq(gh); 1219 if (error) 1220 return error; 1221 1222 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1223 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1224 return 0; 1225 1226 curtime = get_seconds(); 1227 if (curtime - ip->i_di.di_atime >= quantum) { 1228 gfs2_glock_dq(gh); 1229 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1230 gh); 1231 error = gfs2_glock_nq(gh); 1232 if (error) 1233 return error; 1234 1235 /* Verify that atime hasn't been updated while we were 1236 trying to get exclusive lock. */ 1237 1238 curtime = get_seconds(); 1239 if (curtime - ip->i_di.di_atime >= quantum) { 1240 struct buffer_head *dibh; 1241 struct gfs2_dinode *di; 1242 1243 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1244 if (error == -EROFS) 1245 return 0; 1246 if (error) 1247 goto fail; 1248 1249 error = gfs2_meta_inode_buffer(ip, &dibh); 1250 if (error) 1251 goto fail_end_trans; 1252 1253 ip->i_di.di_atime = curtime; 1254 1255 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1256 di = (struct gfs2_dinode *)dibh->b_data; 1257 di->di_atime = cpu_to_be64(ip->i_di.di_atime); 1258 brelse(dibh); 1259 1260 gfs2_trans_end(sdp); 1261 } 1262 1263 /* If someone else has asked for the glock, 1264 unlock and let them have it. Then reacquire 1265 in the original state. */ 1266 if (gfs2_glock_is_blocking(gl)) { 1267 gfs2_glock_dq(gh); 1268 gfs2_holder_reinit(state, flags, gh); 1269 return gfs2_glock_nq(gh); 1270 } 1271 } 1272 1273 return 0; 1274 1275 fail_end_trans: 1276 gfs2_trans_end(sdp); 1277 fail: 1278 gfs2_glock_dq(gh); 1279 return error; 1280 } 1281 1282 /** 1283 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1284 * @arg_a: the first structure 1285 * @arg_b: the second structure 1286 * 1287 * Returns: 1 if A > B 1288 * -1 if A < B 1289 * 0 if A == B 1290 */ 1291 1292 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1293 { 1294 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; 1295 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; 1296 const struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1297 const struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1298 1299 if (a->ln_number > b->ln_number) 1300 return 1; 1301 if (a->ln_number < b->ln_number) 1302 return -1; 1303 if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE) 1304 return 1; 1305 if (gh_a->gh_state == LM_ST_SHARED && (gh_b->gh_flags & GL_ATIME)) 1306 return 1; 1307 1308 return 0; 1309 } 1310 1311 /** 1312 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1313 * atime update 1314 * @num_gh: the number of structures 1315 * @ghs: an array of struct gfs2_holder structures 1316 * 1317 * Returns: 0 on success (all glocks acquired), 1318 * errno on failure (no glocks acquired) 1319 */ 1320 1321 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1322 { 1323 struct gfs2_holder **p; 1324 unsigned int x; 1325 int error = 0; 1326 1327 if (!num_gh) 1328 return 0; 1329 1330 if (num_gh == 1) { 1331 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1332 if (ghs->gh_flags & GL_ATIME) 1333 error = gfs2_glock_nq_atime(ghs); 1334 else 1335 error = gfs2_glock_nq(ghs); 1336 return error; 1337 } 1338 1339 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1340 if (!p) 1341 return -ENOMEM; 1342 1343 for (x = 0; x < num_gh; x++) 1344 p[x] = &ghs[x]; 1345 1346 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1347 1348 for (x = 0; x < num_gh; x++) { 1349 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1350 1351 if (p[x]->gh_flags & GL_ATIME) 1352 error = gfs2_glock_nq_atime(p[x]); 1353 else 1354 error = gfs2_glock_nq(p[x]); 1355 1356 if (error) { 1357 while (x--) 1358 gfs2_glock_dq(p[x]); 1359 break; 1360 } 1361 } 1362 1363 kfree(p); 1364 return error; 1365 } 1366 1367 1368 static int 1369 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1370 { 1371 struct buffer_head *dibh; 1372 int error; 1373 1374 error = gfs2_meta_inode_buffer(ip, &dibh); 1375 if (!error) { 1376 error = inode_setattr(&ip->i_inode, attr); 1377 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1378 gfs2_inode_attr_out(ip); 1379 1380 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1381 gfs2_dinode_out(ip, dibh->b_data); 1382 brelse(dibh); 1383 } 1384 return error; 1385 } 1386 1387 /** 1388 * gfs2_setattr_simple - 1389 * @ip: 1390 * @attr: 1391 * 1392 * Called with a reference on the vnode. 1393 * 1394 * Returns: errno 1395 */ 1396 1397 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1398 { 1399 int error; 1400 1401 if (current->journal_info) 1402 return __gfs2_setattr_simple(ip, attr); 1403 1404 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1405 if (error) 1406 return error; 1407 1408 error = __gfs2_setattr_simple(ip, attr); 1409 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1410 return error; 1411 } 1412 1413