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 55 switch (di->di_mode & S_IFMT) { 56 case S_IFBLK: 57 case S_IFCHR: 58 inode->i_rdev = MKDEV(di->di_major, di->di_minor); 59 break; 60 default: 61 inode->i_rdev = 0; 62 break; 63 }; 64 65 inode->i_mode = di->di_mode; 66 inode->i_nlink = di->di_nlink; 67 inode->i_uid = di->di_uid; 68 inode->i_gid = di->di_gid; 69 i_size_write(inode, di->di_size); 70 inode->i_atime.tv_sec = di->di_atime; 71 inode->i_mtime.tv_sec = di->di_mtime; 72 inode->i_ctime.tv_sec = di->di_ctime; 73 inode->i_atime.tv_nsec = 0; 74 inode->i_mtime.tv_nsec = 0; 75 inode->i_ctime.tv_nsec = 0; 76 inode->i_blocks = di->di_blocks << 77 (GFS2_SB(inode)->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT); 78 79 if (di->di_flags & GFS2_DIF_IMMUTABLE) 80 inode->i_flags |= S_IMMUTABLE; 81 else 82 inode->i_flags &= ~S_IMMUTABLE; 83 84 if (di->di_flags & GFS2_DIF_APPENDONLY) 85 inode->i_flags |= S_APPEND; 86 else 87 inode->i_flags &= ~S_APPEND; 88 } 89 90 /** 91 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode 92 * @ip: The GFS2 inode 93 * 94 * Only copy out the attributes that we want the VFS layer 95 * to be able to modify. 96 */ 97 98 void gfs2_inode_attr_out(struct gfs2_inode *ip) 99 { 100 struct inode *inode = &ip->i_inode; 101 struct gfs2_dinode_host *di = &ip->i_di; 102 gfs2_assert_withdraw(GFS2_SB(inode), 103 (di->di_mode & S_IFMT) == (inode->i_mode & S_IFMT)); 104 di->di_mode = inode->i_mode; 105 di->di_uid = inode->i_uid; 106 di->di_gid = inode->i_gid; 107 di->di_atime = inode->i_atime.tv_sec; 108 di->di_mtime = inode->i_mtime.tv_sec; 109 di->di_ctime = inode->i_ctime.tv_sec; 110 } 111 112 static int iget_test(struct inode *inode, void *opaque) 113 { 114 struct gfs2_inode *ip = GFS2_I(inode); 115 struct gfs2_inum_host *inum = opaque; 116 117 if (ip && ip->i_num.no_addr == inum->no_addr) 118 return 1; 119 120 return 0; 121 } 122 123 static int iget_set(struct inode *inode, void *opaque) 124 { 125 struct gfs2_inode *ip = GFS2_I(inode); 126 struct gfs2_inum_host *inum = opaque; 127 128 ip->i_num = *inum; 129 return 0; 130 } 131 132 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum_host *inum) 133 { 134 return ilookup5(sb, (unsigned long)inum->no_formal_ino, 135 iget_test, inum); 136 } 137 138 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum_host *inum) 139 { 140 return iget5_locked(sb, (unsigned long)inum->no_formal_ino, 141 iget_test, iget_set, inum); 142 } 143 144 /** 145 * gfs2_inode_lookup - Lookup an inode 146 * @sb: The super block 147 * @inum: The inode number 148 * @type: The type of the inode 149 * 150 * Returns: A VFS inode, or an error 151 */ 152 153 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum_host *inum, unsigned int type) 154 { 155 struct inode *inode = gfs2_iget(sb, inum); 156 struct gfs2_inode *ip = GFS2_I(inode); 157 struct gfs2_glock *io_gl; 158 int error; 159 160 if (!inode) 161 return ERR_PTR(-ENOBUFS); 162 163 if (inode->i_state & I_NEW) { 164 struct gfs2_sbd *sdp = GFS2_SB(inode); 165 umode_t mode = DT2IF(type); 166 inode->i_private = ip; 167 inode->i_mode = mode; 168 169 if (S_ISREG(mode)) { 170 inode->i_op = &gfs2_file_iops; 171 inode->i_fop = &gfs2_file_fops; 172 inode->i_mapping->a_ops = &gfs2_file_aops; 173 } else if (S_ISDIR(mode)) { 174 inode->i_op = &gfs2_dir_iops; 175 inode->i_fop = &gfs2_dir_fops; 176 } else if (S_ISLNK(mode)) { 177 inode->i_op = &gfs2_symlink_iops; 178 } else { 179 inode->i_op = &gfs2_dev_iops; 180 } 181 182 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl); 183 if (unlikely(error)) 184 goto fail; 185 ip->i_gl->gl_object = ip; 186 187 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl); 188 if (unlikely(error)) 189 goto fail_put; 190 191 ip->i_vn = ip->i_gl->gl_vn - 1; 192 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh); 193 if (unlikely(error)) 194 goto fail_iopen; 195 196 gfs2_glock_put(io_gl); 197 unlock_new_inode(inode); 198 } 199 200 return inode; 201 fail_iopen: 202 gfs2_glock_put(io_gl); 203 fail_put: 204 ip->i_gl->gl_object = NULL; 205 gfs2_glock_put(ip->i_gl); 206 fail: 207 iput(inode); 208 return ERR_PTR(error); 209 } 210 211 static void gfs2_dinode_in(struct gfs2_inode *ip, const void *buf) 212 { 213 struct gfs2_dinode_host *di = &ip->i_di; 214 const struct gfs2_dinode *str = buf; 215 216 gfs2_meta_header_in(&di->di_header, buf); 217 gfs2_inum_in(&di->di_num, &str->di_num); 218 219 di->di_mode = be32_to_cpu(str->di_mode); 220 di->di_uid = be32_to_cpu(str->di_uid); 221 di->di_gid = be32_to_cpu(str->di_gid); 222 di->di_nlink = be32_to_cpu(str->di_nlink); 223 di->di_size = be64_to_cpu(str->di_size); 224 di->di_blocks = be64_to_cpu(str->di_blocks); 225 di->di_atime = be64_to_cpu(str->di_atime); 226 di->di_mtime = be64_to_cpu(str->di_mtime); 227 di->di_ctime = be64_to_cpu(str->di_ctime); 228 di->di_major = be32_to_cpu(str->di_major); 229 di->di_minor = be32_to_cpu(str->di_minor); 230 231 di->di_goal_meta = be64_to_cpu(str->di_goal_meta); 232 di->di_goal_data = be64_to_cpu(str->di_goal_data); 233 di->di_generation = be64_to_cpu(str->di_generation); 234 235 di->di_flags = be32_to_cpu(str->di_flags); 236 di->di_payload_format = be32_to_cpu(str->di_payload_format); 237 di->di_height = be16_to_cpu(str->di_height); 238 239 di->di_depth = be16_to_cpu(str->di_depth); 240 di->di_entries = be32_to_cpu(str->di_entries); 241 242 di->di_eattr = be64_to_cpu(str->di_eattr); 243 } 244 245 /** 246 * gfs2_inode_refresh - Refresh the incore copy of the dinode 247 * @ip: The GFS2 inode 248 * 249 * Returns: errno 250 */ 251 252 int gfs2_inode_refresh(struct gfs2_inode *ip) 253 { 254 struct buffer_head *dibh; 255 int error; 256 257 error = gfs2_meta_inode_buffer(ip, &dibh); 258 if (error) 259 return error; 260 261 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) { 262 brelse(dibh); 263 return -EIO; 264 } 265 266 gfs2_dinode_in(ip, dibh->b_data); 267 268 brelse(dibh); 269 270 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) { 271 if (gfs2_consist_inode(ip)) 272 gfs2_dinode_print(&ip->i_di); 273 return -EIO; 274 } 275 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino) 276 return -ESTALE; 277 278 ip->i_vn = ip->i_gl->gl_vn; 279 280 return 0; 281 } 282 283 int gfs2_dinode_dealloc(struct gfs2_inode *ip) 284 { 285 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 286 struct gfs2_alloc *al; 287 struct gfs2_rgrpd *rgd; 288 int error; 289 290 if (ip->i_di.di_blocks != 1) { 291 if (gfs2_consist_inode(ip)) 292 gfs2_dinode_print(&ip->i_di); 293 return -EIO; 294 } 295 296 al = gfs2_alloc_get(ip); 297 298 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 299 if (error) 300 goto out; 301 302 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 303 if (error) 304 goto out_qs; 305 306 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 307 if (!rgd) { 308 gfs2_consist_inode(ip); 309 error = -EIO; 310 goto out_rindex_relse; 311 } 312 313 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 314 &al->al_rgd_gh); 315 if (error) 316 goto out_rindex_relse; 317 318 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1); 319 if (error) 320 goto out_rg_gunlock; 321 322 gfs2_trans_add_gl(ip->i_gl); 323 324 gfs2_free_di(rgd, ip); 325 326 gfs2_trans_end(sdp); 327 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 328 329 out_rg_gunlock: 330 gfs2_glock_dq_uninit(&al->al_rgd_gh); 331 out_rindex_relse: 332 gfs2_glock_dq_uninit(&al->al_ri_gh); 333 out_qs: 334 gfs2_quota_unhold(ip); 335 out: 336 gfs2_alloc_put(ip); 337 return error; 338 } 339 340 /** 341 * gfs2_change_nlink - Change nlink count on inode 342 * @ip: The GFS2 inode 343 * @diff: The change in the nlink count required 344 * 345 * Returns: errno 346 */ 347 348 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 349 { 350 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info; 351 struct buffer_head *dibh; 352 u32 nlink; 353 int error; 354 355 BUG_ON(ip->i_di.di_nlink != ip->i_inode.i_nlink); 356 nlink = ip->i_di.di_nlink + diff; 357 358 /* If we are reducing the nlink count, but the new value ends up being 359 bigger than the old one, we must have underflowed. */ 360 if (diff < 0 && nlink > ip->i_di.di_nlink) { 361 if (gfs2_consist_inode(ip)) 362 gfs2_dinode_print(&ip->i_di); 363 return -EIO; 364 } 365 366 error = gfs2_meta_inode_buffer(ip, &dibh); 367 if (error) 368 return error; 369 370 ip->i_di.di_nlink = nlink; 371 ip->i_di.di_ctime = get_seconds(); 372 ip->i_inode.i_nlink = nlink; 373 374 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 375 gfs2_dinode_out(ip, dibh->b_data); 376 brelse(dibh); 377 mark_inode_dirty(&ip->i_inode); 378 379 if (ip->i_di.di_nlink == 0) { 380 struct gfs2_rgrpd *rgd; 381 struct gfs2_holder ri_gh, rg_gh; 382 383 error = gfs2_rindex_hold(sdp, &ri_gh); 384 if (error) 385 goto out; 386 error = -EIO; 387 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 388 if (!rgd) 389 goto out_norgrp; 390 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh); 391 if (error) 392 goto out_norgrp; 393 394 clear_nlink(&ip->i_inode); 395 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 396 gfs2_glock_dq_uninit(&rg_gh); 397 out_norgrp: 398 gfs2_glock_dq_uninit(&ri_gh); 399 } 400 out: 401 return error; 402 } 403 404 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 405 { 406 struct qstr qstr; 407 gfs2_str2qstr(&qstr, name); 408 return gfs2_lookupi(dip, &qstr, 1, NULL); 409 } 410 411 412 /** 413 * gfs2_lookupi - Look up a filename in a directory and return its inode 414 * @d_gh: An initialized holder for the directory glock 415 * @name: The name of the inode to look for 416 * @is_root: If 1, ignore the caller's permissions 417 * @i_gh: An uninitialized holder for the new inode glock 418 * 419 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 420 * @is_root is true. 421 * 422 * Returns: errno 423 */ 424 425 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 426 int is_root, struct nameidata *nd) 427 { 428 struct super_block *sb = dir->i_sb; 429 struct gfs2_inode *dip = GFS2_I(dir); 430 struct gfs2_holder d_gh; 431 struct gfs2_inum_host inum; 432 unsigned int type; 433 int error = 0; 434 struct inode *inode = NULL; 435 436 if (!name->len || name->len > GFS2_FNAMESIZE) 437 return ERR_PTR(-ENAMETOOLONG); 438 439 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 440 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 441 dir == sb->s_root->d_inode)) { 442 igrab(dir); 443 return dir; 444 } 445 446 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 447 if (error) 448 return ERR_PTR(error); 449 450 if (!is_root) { 451 error = permission(dir, MAY_EXEC, NULL); 452 if (error) 453 goto out; 454 } 455 456 error = gfs2_dir_search(dir, name, &inum, &type); 457 if (error) 458 goto out; 459 460 inode = gfs2_inode_lookup(sb, &inum, type); 461 462 out: 463 gfs2_glock_dq_uninit(&d_gh); 464 if (error == -ENOENT) 465 return NULL; 466 return inode; 467 } 468 469 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino) 470 { 471 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 472 struct buffer_head *bh; 473 struct gfs2_inum_range_host ir; 474 int error; 475 476 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 477 if (error) 478 return error; 479 mutex_lock(&sdp->sd_inum_mutex); 480 481 error = gfs2_meta_inode_buffer(ip, &bh); 482 if (error) { 483 mutex_unlock(&sdp->sd_inum_mutex); 484 gfs2_trans_end(sdp); 485 return error; 486 } 487 488 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 489 490 if (ir.ir_length) { 491 *formal_ino = ir.ir_start++; 492 ir.ir_length--; 493 gfs2_trans_add_bh(ip->i_gl, bh, 1); 494 gfs2_inum_range_out(&ir, 495 bh->b_data + sizeof(struct gfs2_dinode)); 496 brelse(bh); 497 mutex_unlock(&sdp->sd_inum_mutex); 498 gfs2_trans_end(sdp); 499 return 0; 500 } 501 502 brelse(bh); 503 504 mutex_unlock(&sdp->sd_inum_mutex); 505 gfs2_trans_end(sdp); 506 507 return 1; 508 } 509 510 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino) 511 { 512 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 513 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 514 struct gfs2_holder gh; 515 struct buffer_head *bh; 516 struct gfs2_inum_range_host ir; 517 int error; 518 519 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 520 if (error) 521 return error; 522 523 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 524 if (error) 525 goto out; 526 mutex_lock(&sdp->sd_inum_mutex); 527 528 error = gfs2_meta_inode_buffer(ip, &bh); 529 if (error) 530 goto out_end_trans; 531 532 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 533 534 if (!ir.ir_length) { 535 struct buffer_head *m_bh; 536 u64 x, y; 537 __be64 z; 538 539 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 540 if (error) 541 goto out_brelse; 542 543 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 544 x = y = be64_to_cpu(z); 545 ir.ir_start = x; 546 ir.ir_length = GFS2_INUM_QUANTUM; 547 x += GFS2_INUM_QUANTUM; 548 if (x < y) 549 gfs2_consist_inode(m_ip); 550 z = cpu_to_be64(x); 551 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 552 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z; 553 554 brelse(m_bh); 555 } 556 557 *formal_ino = ir.ir_start++; 558 ir.ir_length--; 559 560 gfs2_trans_add_bh(ip->i_gl, bh, 1); 561 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 562 563 out_brelse: 564 brelse(bh); 565 out_end_trans: 566 mutex_unlock(&sdp->sd_inum_mutex); 567 gfs2_trans_end(sdp); 568 out: 569 gfs2_glock_dq_uninit(&gh); 570 return error; 571 } 572 573 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum) 574 { 575 int error; 576 577 error = pick_formal_ino_1(sdp, inum); 578 if (error <= 0) 579 return error; 580 581 error = pick_formal_ino_2(sdp, inum); 582 583 return error; 584 } 585 586 /** 587 * create_ok - OK to create a new on-disk inode here? 588 * @dip: Directory in which dinode is to be created 589 * @name: Name of new dinode 590 * @mode: 591 * 592 * Returns: errno 593 */ 594 595 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 596 unsigned int mode) 597 { 598 int error; 599 600 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 601 if (error) 602 return error; 603 604 /* Don't create entries in an unlinked directory */ 605 if (!dip->i_di.di_nlink) 606 return -EPERM; 607 608 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL); 609 switch (error) { 610 case -ENOENT: 611 error = 0; 612 break; 613 case 0: 614 return -EEXIST; 615 default: 616 return error; 617 } 618 619 if (dip->i_di.di_entries == (u32)-1) 620 return -EFBIG; 621 if (S_ISDIR(mode) && dip->i_di.di_nlink == (u32)-1) 622 return -EMLINK; 623 624 return 0; 625 } 626 627 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 628 unsigned int *uid, unsigned int *gid) 629 { 630 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 631 (dip->i_di.di_mode & S_ISUID) && dip->i_di.di_uid) { 632 if (S_ISDIR(*mode)) 633 *mode |= S_ISUID; 634 else if (dip->i_di.di_uid != current->fsuid) 635 *mode &= ~07111; 636 *uid = dip->i_di.di_uid; 637 } else 638 *uid = current->fsuid; 639 640 if (dip->i_di.di_mode & S_ISGID) { 641 if (S_ISDIR(*mode)) 642 *mode |= S_ISGID; 643 *gid = dip->i_di.di_gid; 644 } else 645 *gid = current->fsgid; 646 } 647 648 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum, 649 u64 *generation) 650 { 651 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 652 int error; 653 654 gfs2_alloc_get(dip); 655 656 dip->i_alloc.al_requested = RES_DINODE; 657 error = gfs2_inplace_reserve(dip); 658 if (error) 659 goto out; 660 661 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 662 if (error) 663 goto out_ipreserv; 664 665 inum->no_addr = gfs2_alloc_di(dip, generation); 666 667 gfs2_trans_end(sdp); 668 669 out_ipreserv: 670 gfs2_inplace_release(dip); 671 out: 672 gfs2_alloc_put(dip); 673 return error; 674 } 675 676 /** 677 * init_dinode - Fill in a new dinode structure 678 * @dip: the directory this inode is being created in 679 * @gl: The glock covering the new inode 680 * @inum: the inode number 681 * @mode: the file permissions 682 * @uid: 683 * @gid: 684 * 685 */ 686 687 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 688 const struct gfs2_inum_host *inum, unsigned int mode, 689 unsigned int uid, unsigned int gid, 690 const u64 *generation) 691 { 692 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 693 struct gfs2_dinode *di; 694 struct buffer_head *dibh; 695 696 dibh = gfs2_meta_new(gl, inum->no_addr); 697 gfs2_trans_add_bh(gl, dibh, 1); 698 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 699 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 700 di = (struct gfs2_dinode *)dibh->b_data; 701 702 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 703 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 704 di->di_mode = cpu_to_be32(mode); 705 di->di_uid = cpu_to_be32(uid); 706 di->di_gid = cpu_to_be32(gid); 707 di->di_nlink = cpu_to_be32(0); 708 di->di_size = cpu_to_be64(0); 709 di->di_blocks = cpu_to_be64(1); 710 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 711 di->di_major = di->di_minor = cpu_to_be32(0); 712 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 713 di->di_generation = cpu_to_be64(*generation); 714 di->di_flags = cpu_to_be32(0); 715 716 if (S_ISREG(mode)) { 717 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 718 gfs2_tune_get(sdp, gt_new_files_jdata)) 719 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 720 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 721 gfs2_tune_get(sdp, gt_new_files_directio)) 722 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 723 } else if (S_ISDIR(mode)) { 724 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 725 GFS2_DIF_INHERIT_DIRECTIO); 726 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 727 GFS2_DIF_INHERIT_JDATA); 728 } 729 730 di->__pad1 = 0; 731 di->di_payload_format = cpu_to_be32(0); 732 di->di_height = cpu_to_be32(0); 733 di->__pad2 = 0; 734 di->__pad3 = 0; 735 di->di_depth = cpu_to_be16(0); 736 di->di_entries = cpu_to_be32(0); 737 memset(&di->__pad4, 0, sizeof(di->__pad4)); 738 di->di_eattr = cpu_to_be64(0); 739 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 740 741 brelse(dibh); 742 } 743 744 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 745 unsigned int mode, const struct gfs2_inum_host *inum, 746 const u64 *generation) 747 { 748 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 749 unsigned int uid, gid; 750 int error; 751 752 munge_mode_uid_gid(dip, &mode, &uid, &gid); 753 gfs2_alloc_get(dip); 754 755 error = gfs2_quota_lock(dip, uid, gid); 756 if (error) 757 goto out; 758 759 error = gfs2_quota_check(dip, uid, gid); 760 if (error) 761 goto out_quota; 762 763 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0); 764 if (error) 765 goto out_quota; 766 767 init_dinode(dip, gl, inum, mode, uid, gid, generation); 768 gfs2_quota_change(dip, +1, uid, gid); 769 gfs2_trans_end(sdp); 770 771 out_quota: 772 gfs2_quota_unlock(dip); 773 out: 774 gfs2_alloc_put(dip); 775 return error; 776 } 777 778 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name, 779 struct gfs2_inode *ip) 780 { 781 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 782 struct gfs2_alloc *al; 783 int alloc_required; 784 struct buffer_head *dibh; 785 int error; 786 787 al = gfs2_alloc_get(dip); 788 789 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 790 if (error) 791 goto fail; 792 793 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name); 794 if (alloc_required < 0) 795 goto fail; 796 if (alloc_required) { 797 error = gfs2_quota_check(dip, dip->i_di.di_uid, 798 dip->i_di.di_gid); 799 if (error) 800 goto fail_quota_locks; 801 802 al->al_requested = sdp->sd_max_dirres; 803 804 error = gfs2_inplace_reserve(dip); 805 if (error) 806 goto fail_quota_locks; 807 808 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 809 al->al_rgd->rd_ri.ri_length + 810 2 * RES_DINODE + 811 RES_STATFS + RES_QUOTA, 0); 812 if (error) 813 goto fail_ipreserv; 814 } else { 815 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 816 if (error) 817 goto fail_quota_locks; 818 } 819 820 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_di.di_mode)); 821 if (error) 822 goto fail_end_trans; 823 824 error = gfs2_meta_inode_buffer(ip, &dibh); 825 if (error) 826 goto fail_end_trans; 827 ip->i_di.di_nlink = 1; 828 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 829 gfs2_dinode_out(ip, dibh->b_data); 830 brelse(dibh); 831 return 0; 832 833 fail_end_trans: 834 gfs2_trans_end(sdp); 835 836 fail_ipreserv: 837 if (dip->i_alloc.al_rgd) 838 gfs2_inplace_release(dip); 839 840 fail_quota_locks: 841 gfs2_quota_unlock(dip); 842 843 fail: 844 gfs2_alloc_put(dip); 845 return error; 846 } 847 848 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip) 849 { 850 int err; 851 size_t len; 852 void *value; 853 char *name; 854 struct gfs2_ea_request er; 855 856 err = security_inode_init_security(&ip->i_inode, &dip->i_inode, 857 &name, &value, &len); 858 859 if (err) { 860 if (err == -EOPNOTSUPP) 861 return 0; 862 return err; 863 } 864 865 memset(&er, 0, sizeof(struct gfs2_ea_request)); 866 867 er.er_type = GFS2_EATYPE_SECURITY; 868 er.er_name = name; 869 er.er_data = value; 870 er.er_name_len = strlen(name); 871 er.er_data_len = len; 872 873 err = gfs2_ea_set_i(ip, &er); 874 875 kfree(value); 876 kfree(name); 877 878 return err; 879 } 880 881 /** 882 * gfs2_createi - Create a new inode 883 * @ghs: An array of two holders 884 * @name: The name of the new file 885 * @mode: the permissions on the new inode 886 * 887 * @ghs[0] is an initialized holder for the directory 888 * @ghs[1] is the holder for the inode lock 889 * 890 * If the return value is not NULL, the glocks on both the directory and the new 891 * file are held. A transaction has been started and an inplace reservation 892 * is held, as well. 893 * 894 * Returns: An inode 895 */ 896 897 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 898 unsigned int mode) 899 { 900 struct inode *inode; 901 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 902 struct inode *dir = &dip->i_inode; 903 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 904 struct gfs2_inum_host inum; 905 int error; 906 u64 generation; 907 908 if (!name->len || name->len > GFS2_FNAMESIZE) 909 return ERR_PTR(-ENAMETOOLONG); 910 911 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 912 error = gfs2_glock_nq(ghs); 913 if (error) 914 goto fail; 915 916 error = create_ok(dip, name, mode); 917 if (error) 918 goto fail_gunlock; 919 920 error = pick_formal_ino(sdp, &inum.no_formal_ino); 921 if (error) 922 goto fail_gunlock; 923 924 error = alloc_dinode(dip, &inum, &generation); 925 if (error) 926 goto fail_gunlock; 927 928 if (inum.no_addr < dip->i_num.no_addr) { 929 gfs2_glock_dq(ghs); 930 931 error = gfs2_glock_nq_num(sdp, inum.no_addr, 932 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 933 GL_SKIP, ghs + 1); 934 if (error) { 935 return ERR_PTR(error); 936 } 937 938 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 939 error = gfs2_glock_nq(ghs); 940 if (error) { 941 gfs2_glock_dq_uninit(ghs + 1); 942 return ERR_PTR(error); 943 } 944 945 error = create_ok(dip, name, mode); 946 if (error) 947 goto fail_gunlock2; 948 } else { 949 error = gfs2_glock_nq_num(sdp, inum.no_addr, 950 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 951 GL_SKIP, ghs + 1); 952 if (error) 953 goto fail_gunlock; 954 } 955 956 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation); 957 if (error) 958 goto fail_gunlock2; 959 960 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode)); 961 if (IS_ERR(inode)) 962 goto fail_gunlock2; 963 964 error = gfs2_inode_refresh(GFS2_I(inode)); 965 if (error) 966 goto fail_iput; 967 968 error = gfs2_acl_create(dip, GFS2_I(inode)); 969 if (error) 970 goto fail_iput; 971 972 error = gfs2_security_init(dip, GFS2_I(inode)); 973 if (error) 974 goto fail_iput; 975 976 error = link_dinode(dip, name, GFS2_I(inode)); 977 if (error) 978 goto fail_iput; 979 980 if (!inode) 981 return ERR_PTR(-ENOMEM); 982 return inode; 983 984 fail_iput: 985 iput(inode); 986 fail_gunlock2: 987 gfs2_glock_dq_uninit(ghs + 1); 988 fail_gunlock: 989 gfs2_glock_dq(ghs); 990 fail: 991 return ERR_PTR(error); 992 } 993 994 /** 995 * gfs2_rmdiri - Remove a directory 996 * @dip: The parent directory of the directory to be removed 997 * @name: The name of the directory to be removed 998 * @ip: The GFS2 inode of the directory to be removed 999 * 1000 * Assumes Glocks on dip and ip are held 1001 * 1002 * Returns: errno 1003 */ 1004 1005 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 1006 struct gfs2_inode *ip) 1007 { 1008 struct qstr dotname; 1009 int error; 1010 1011 if (ip->i_di.di_entries != 2) { 1012 if (gfs2_consist_inode(ip)) 1013 gfs2_dinode_print(&ip->i_di); 1014 return -EIO; 1015 } 1016 1017 error = gfs2_dir_del(dip, name); 1018 if (error) 1019 return error; 1020 1021 error = gfs2_change_nlink(dip, -1); 1022 if (error) 1023 return error; 1024 1025 gfs2_str2qstr(&dotname, "."); 1026 error = gfs2_dir_del(ip, &dotname); 1027 if (error) 1028 return error; 1029 1030 gfs2_str2qstr(&dotname, ".."); 1031 error = gfs2_dir_del(ip, &dotname); 1032 if (error) 1033 return error; 1034 1035 error = gfs2_change_nlink(ip, -2); 1036 if (error) 1037 return error; 1038 1039 return error; 1040 } 1041 1042 /* 1043 * gfs2_unlink_ok - check to see that a inode is still in a directory 1044 * @dip: the directory 1045 * @name: the name of the file 1046 * @ip: the inode 1047 * 1048 * Assumes that the lock on (at least) @dip is held. 1049 * 1050 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 1051 */ 1052 1053 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 1054 struct gfs2_inode *ip) 1055 { 1056 struct gfs2_inum_host inum; 1057 unsigned int type; 1058 int error; 1059 1060 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 1061 return -EPERM; 1062 1063 if ((dip->i_di.di_mode & S_ISVTX) && 1064 dip->i_di.di_uid != current->fsuid && 1065 ip->i_di.di_uid != current->fsuid && !capable(CAP_FOWNER)) 1066 return -EPERM; 1067 1068 if (IS_APPEND(&dip->i_inode)) 1069 return -EPERM; 1070 1071 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 1072 if (error) 1073 return error; 1074 1075 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type); 1076 if (error) 1077 return error; 1078 1079 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1080 return -ENOENT; 1081 1082 if (IF2DT(ip->i_di.di_mode) != type) { 1083 gfs2_consist_inode(dip); 1084 return -EIO; 1085 } 1086 1087 return 0; 1088 } 1089 1090 /* 1091 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1092 * @this: move this 1093 * @to: to here 1094 * 1095 * Follow @to back to the root and make sure we don't encounter @this 1096 * Assumes we already hold the rename lock. 1097 * 1098 * Returns: errno 1099 */ 1100 1101 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1102 { 1103 struct inode *dir = &to->i_inode; 1104 struct super_block *sb = dir->i_sb; 1105 struct inode *tmp; 1106 struct qstr dotdot; 1107 int error = 0; 1108 1109 gfs2_str2qstr(&dotdot, ".."); 1110 1111 igrab(dir); 1112 1113 for (;;) { 1114 if (dir == &this->i_inode) { 1115 error = -EINVAL; 1116 break; 1117 } 1118 if (dir == sb->s_root->d_inode) { 1119 error = 0; 1120 break; 1121 } 1122 1123 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1124 if (IS_ERR(tmp)) { 1125 error = PTR_ERR(tmp); 1126 break; 1127 } 1128 1129 iput(dir); 1130 dir = tmp; 1131 } 1132 1133 iput(dir); 1134 1135 return error; 1136 } 1137 1138 /** 1139 * gfs2_readlinki - return the contents of a symlink 1140 * @ip: the symlink's inode 1141 * @buf: a pointer to the buffer to be filled 1142 * @len: a pointer to the length of @buf 1143 * 1144 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1145 * to be freed by the caller. 1146 * 1147 * Returns: errno 1148 */ 1149 1150 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1151 { 1152 struct gfs2_holder i_gh; 1153 struct buffer_head *dibh; 1154 unsigned int x; 1155 int error; 1156 1157 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1158 error = gfs2_glock_nq_atime(&i_gh); 1159 if (error) { 1160 gfs2_holder_uninit(&i_gh); 1161 return error; 1162 } 1163 1164 if (!ip->i_di.di_size) { 1165 gfs2_consist_inode(ip); 1166 error = -EIO; 1167 goto out; 1168 } 1169 1170 error = gfs2_meta_inode_buffer(ip, &dibh); 1171 if (error) 1172 goto out; 1173 1174 x = ip->i_di.di_size + 1; 1175 if (x > *len) { 1176 *buf = kmalloc(x, GFP_KERNEL); 1177 if (!*buf) { 1178 error = -ENOMEM; 1179 goto out_brelse; 1180 } 1181 } 1182 1183 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1184 *len = x; 1185 1186 out_brelse: 1187 brelse(dibh); 1188 out: 1189 gfs2_glock_dq_uninit(&i_gh); 1190 return error; 1191 } 1192 1193 /** 1194 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1195 * conditionally update the inode's atime 1196 * @gh: the holder to acquire 1197 * 1198 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1199 * Update if the difference between the current time and the inode's current 1200 * atime is greater than an interval specified at mount. 1201 * 1202 * Returns: errno 1203 */ 1204 1205 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1206 { 1207 struct gfs2_glock *gl = gh->gh_gl; 1208 struct gfs2_sbd *sdp = gl->gl_sbd; 1209 struct gfs2_inode *ip = gl->gl_object; 1210 s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1211 unsigned int state; 1212 int flags; 1213 int error; 1214 1215 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1216 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1217 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1218 return -EINVAL; 1219 1220 state = gh->gh_state; 1221 flags = gh->gh_flags; 1222 1223 error = gfs2_glock_nq(gh); 1224 if (error) 1225 return error; 1226 1227 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1228 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1229 return 0; 1230 1231 curtime = get_seconds(); 1232 if (curtime - ip->i_di.di_atime >= quantum) { 1233 gfs2_glock_dq(gh); 1234 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1235 gh); 1236 error = gfs2_glock_nq(gh); 1237 if (error) 1238 return error; 1239 1240 /* Verify that atime hasn't been updated while we were 1241 trying to get exclusive lock. */ 1242 1243 curtime = get_seconds(); 1244 if (curtime - ip->i_di.di_atime >= quantum) { 1245 struct buffer_head *dibh; 1246 struct gfs2_dinode *di; 1247 1248 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1249 if (error == -EROFS) 1250 return 0; 1251 if (error) 1252 goto fail; 1253 1254 error = gfs2_meta_inode_buffer(ip, &dibh); 1255 if (error) 1256 goto fail_end_trans; 1257 1258 ip->i_di.di_atime = curtime; 1259 1260 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1261 di = (struct gfs2_dinode *)dibh->b_data; 1262 di->di_atime = cpu_to_be64(ip->i_di.di_atime); 1263 brelse(dibh); 1264 1265 gfs2_trans_end(sdp); 1266 } 1267 1268 /* If someone else has asked for the glock, 1269 unlock and let them have it. Then reacquire 1270 in the original state. */ 1271 if (gfs2_glock_is_blocking(gl)) { 1272 gfs2_glock_dq(gh); 1273 gfs2_holder_reinit(state, flags, gh); 1274 return gfs2_glock_nq(gh); 1275 } 1276 } 1277 1278 return 0; 1279 1280 fail_end_trans: 1281 gfs2_trans_end(sdp); 1282 fail: 1283 gfs2_glock_dq(gh); 1284 return error; 1285 } 1286 1287 /** 1288 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1289 * @arg_a: the first structure 1290 * @arg_b: the second structure 1291 * 1292 * Returns: 1 if A > B 1293 * -1 if A < B 1294 * 0 if A == B 1295 */ 1296 1297 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1298 { 1299 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; 1300 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; 1301 const struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1302 const struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1303 1304 if (a->ln_number > b->ln_number) 1305 return 1; 1306 if (a->ln_number < b->ln_number) 1307 return -1; 1308 if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE) 1309 return 1; 1310 if (gh_a->gh_state == LM_ST_SHARED && (gh_b->gh_flags & GL_ATIME)) 1311 return 1; 1312 1313 return 0; 1314 } 1315 1316 /** 1317 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1318 * atime update 1319 * @num_gh: the number of structures 1320 * @ghs: an array of struct gfs2_holder structures 1321 * 1322 * Returns: 0 on success (all glocks acquired), 1323 * errno on failure (no glocks acquired) 1324 */ 1325 1326 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1327 { 1328 struct gfs2_holder **p; 1329 unsigned int x; 1330 int error = 0; 1331 1332 if (!num_gh) 1333 return 0; 1334 1335 if (num_gh == 1) { 1336 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1337 if (ghs->gh_flags & GL_ATIME) 1338 error = gfs2_glock_nq_atime(ghs); 1339 else 1340 error = gfs2_glock_nq(ghs); 1341 return error; 1342 } 1343 1344 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1345 if (!p) 1346 return -ENOMEM; 1347 1348 for (x = 0; x < num_gh; x++) 1349 p[x] = &ghs[x]; 1350 1351 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1352 1353 for (x = 0; x < num_gh; x++) { 1354 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1355 1356 if (p[x]->gh_flags & GL_ATIME) 1357 error = gfs2_glock_nq_atime(p[x]); 1358 else 1359 error = gfs2_glock_nq(p[x]); 1360 1361 if (error) { 1362 while (x--) 1363 gfs2_glock_dq(p[x]); 1364 break; 1365 } 1366 } 1367 1368 kfree(p); 1369 return error; 1370 } 1371 1372 1373 static int 1374 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1375 { 1376 struct buffer_head *dibh; 1377 int error; 1378 1379 error = gfs2_meta_inode_buffer(ip, &dibh); 1380 if (!error) { 1381 error = inode_setattr(&ip->i_inode, attr); 1382 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1383 gfs2_inode_attr_out(ip); 1384 1385 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1386 gfs2_dinode_out(ip, dibh->b_data); 1387 brelse(dibh); 1388 } 1389 return error; 1390 } 1391 1392 /** 1393 * gfs2_setattr_simple - 1394 * @ip: 1395 * @attr: 1396 * 1397 * Called with a reference on the vnode. 1398 * 1399 * Returns: errno 1400 */ 1401 1402 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1403 { 1404 int error; 1405 1406 if (current->journal_info) 1407 return __gfs2_setattr_simple(ip, attr); 1408 1409 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1410 if (error) 1411 return error; 1412 1413 error = __gfs2_setattr_simple(ip, attr); 1414 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1415 return error; 1416 } 1417 1418