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 i_size_write(inode, di->di_size); 54 inode->i_blocks = di->di_blocks << 55 (GFS2_SB(inode)->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT); 56 } 57 58 static int iget_test(struct inode *inode, void *opaque) 59 { 60 struct gfs2_inode *ip = GFS2_I(inode); 61 struct gfs2_inum_host *inum = opaque; 62 63 if (ip && ip->i_num.no_addr == inum->no_addr) 64 return 1; 65 66 return 0; 67 } 68 69 static int iget_set(struct inode *inode, void *opaque) 70 { 71 struct gfs2_inode *ip = GFS2_I(inode); 72 struct gfs2_inum_host *inum = opaque; 73 74 ip->i_num = *inum; 75 inode->i_ino = inum->no_addr; 76 return 0; 77 } 78 79 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum_host *inum) 80 { 81 return ilookup5(sb, (unsigned long)inum->no_formal_ino, 82 iget_test, inum); 83 } 84 85 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum_host *inum) 86 { 87 return iget5_locked(sb, (unsigned long)inum->no_formal_ino, 88 iget_test, iget_set, inum); 89 } 90 91 /** 92 * gfs2_inode_lookup - Lookup an inode 93 * @sb: The super block 94 * @inum: The inode number 95 * @type: The type of the inode 96 * 97 * Returns: A VFS inode, or an error 98 */ 99 100 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum_host *inum, unsigned int type) 101 { 102 struct inode *inode = gfs2_iget(sb, inum); 103 struct gfs2_inode *ip = GFS2_I(inode); 104 struct gfs2_glock *io_gl; 105 int error; 106 107 if (!inode) 108 return ERR_PTR(-ENOBUFS); 109 110 if (inode->i_state & I_NEW) { 111 struct gfs2_sbd *sdp = GFS2_SB(inode); 112 umode_t mode = DT2IF(type); 113 inode->i_private = ip; 114 inode->i_mode = mode; 115 116 if (S_ISREG(mode)) { 117 inode->i_op = &gfs2_file_iops; 118 inode->i_fop = &gfs2_file_fops; 119 inode->i_mapping->a_ops = &gfs2_file_aops; 120 } else if (S_ISDIR(mode)) { 121 inode->i_op = &gfs2_dir_iops; 122 inode->i_fop = &gfs2_dir_fops; 123 } else if (S_ISLNK(mode)) { 124 inode->i_op = &gfs2_symlink_iops; 125 } else { 126 inode->i_op = &gfs2_dev_iops; 127 } 128 129 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl); 130 if (unlikely(error)) 131 goto fail; 132 ip->i_gl->gl_object = ip; 133 134 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl); 135 if (unlikely(error)) 136 goto fail_put; 137 138 set_bit(GIF_INVALID, &ip->i_flags); 139 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh); 140 if (unlikely(error)) 141 goto fail_iopen; 142 143 gfs2_glock_put(io_gl); 144 unlock_new_inode(inode); 145 } 146 147 return inode; 148 fail_iopen: 149 gfs2_glock_put(io_gl); 150 fail_put: 151 ip->i_gl->gl_object = NULL; 152 gfs2_glock_put(ip->i_gl); 153 fail: 154 iput(inode); 155 return ERR_PTR(error); 156 } 157 158 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf) 159 { 160 struct gfs2_dinode_host *di = &ip->i_di; 161 const struct gfs2_dinode *str = buf; 162 163 if (ip->i_num.no_addr != be64_to_cpu(str->di_num.no_addr)) { 164 if (gfs2_consist_inode(ip)) 165 gfs2_dinode_print(ip); 166 return -EIO; 167 } 168 if (ip->i_num.no_formal_ino != be64_to_cpu(str->di_num.no_formal_ino)) 169 return -ESTALE; 170 171 ip->i_inode.i_mode = be32_to_cpu(str->di_mode); 172 ip->i_inode.i_rdev = 0; 173 switch (ip->i_inode.i_mode & S_IFMT) { 174 case S_IFBLK: 175 case S_IFCHR: 176 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major), 177 be32_to_cpu(str->di_minor)); 178 break; 179 }; 180 181 ip->i_inode.i_uid = be32_to_cpu(str->di_uid); 182 ip->i_inode.i_gid = be32_to_cpu(str->di_gid); 183 /* 184 * We will need to review setting the nlink count here in the 185 * light of the forthcoming ro bind mount work. This is a reminder 186 * to do that. 187 */ 188 ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink); 189 di->di_size = be64_to_cpu(str->di_size); 190 di->di_blocks = be64_to_cpu(str->di_blocks); 191 ip->i_inode.i_atime.tv_sec = be64_to_cpu(str->di_atime); 192 ip->i_inode.i_atime.tv_nsec = 0; 193 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime); 194 ip->i_inode.i_mtime.tv_nsec = 0; 195 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime); 196 ip->i_inode.i_ctime.tv_nsec = 0; 197 198 di->di_goal_meta = be64_to_cpu(str->di_goal_meta); 199 di->di_goal_data = be64_to_cpu(str->di_goal_data); 200 di->di_generation = be64_to_cpu(str->di_generation); 201 202 di->di_flags = be32_to_cpu(str->di_flags); 203 gfs2_set_inode_flags(&ip->i_inode); 204 di->di_height = be16_to_cpu(str->di_height); 205 206 di->di_depth = be16_to_cpu(str->di_depth); 207 di->di_entries = be32_to_cpu(str->di_entries); 208 209 di->di_eattr = be64_to_cpu(str->di_eattr); 210 return 0; 211 } 212 213 /** 214 * gfs2_inode_refresh - Refresh the incore copy of the dinode 215 * @ip: The GFS2 inode 216 * 217 * Returns: errno 218 */ 219 220 int gfs2_inode_refresh(struct gfs2_inode *ip) 221 { 222 struct buffer_head *dibh; 223 int error; 224 225 error = gfs2_meta_inode_buffer(ip, &dibh); 226 if (error) 227 return error; 228 229 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) { 230 brelse(dibh); 231 return -EIO; 232 } 233 234 error = gfs2_dinode_in(ip, dibh->b_data); 235 brelse(dibh); 236 clear_bit(GIF_INVALID, &ip->i_flags); 237 238 return error; 239 } 240 241 int gfs2_dinode_dealloc(struct gfs2_inode *ip) 242 { 243 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 244 struct gfs2_alloc *al; 245 struct gfs2_rgrpd *rgd; 246 int error; 247 248 if (ip->i_di.di_blocks != 1) { 249 if (gfs2_consist_inode(ip)) 250 gfs2_dinode_print(ip); 251 return -EIO; 252 } 253 254 al = gfs2_alloc_get(ip); 255 256 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 257 if (error) 258 goto out; 259 260 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 261 if (error) 262 goto out_qs; 263 264 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 265 if (!rgd) { 266 gfs2_consist_inode(ip); 267 error = -EIO; 268 goto out_rindex_relse; 269 } 270 271 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 272 &al->al_rgd_gh); 273 if (error) 274 goto out_rindex_relse; 275 276 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1); 277 if (error) 278 goto out_rg_gunlock; 279 280 gfs2_trans_add_gl(ip->i_gl); 281 282 gfs2_free_di(rgd, ip); 283 284 gfs2_trans_end(sdp); 285 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 286 287 out_rg_gunlock: 288 gfs2_glock_dq_uninit(&al->al_rgd_gh); 289 out_rindex_relse: 290 gfs2_glock_dq_uninit(&al->al_ri_gh); 291 out_qs: 292 gfs2_quota_unhold(ip); 293 out: 294 gfs2_alloc_put(ip); 295 return error; 296 } 297 298 /** 299 * gfs2_change_nlink - Change nlink count on inode 300 * @ip: The GFS2 inode 301 * @diff: The change in the nlink count required 302 * 303 * Returns: errno 304 */ 305 306 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 307 { 308 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info; 309 struct buffer_head *dibh; 310 u32 nlink; 311 int error; 312 313 BUG_ON(diff != 1 && diff != -1); 314 nlink = ip->i_inode.i_nlink + diff; 315 316 /* If we are reducing the nlink count, but the new value ends up being 317 bigger than the old one, we must have underflowed. */ 318 if (diff < 0 && nlink > ip->i_inode.i_nlink) { 319 if (gfs2_consist_inode(ip)) 320 gfs2_dinode_print(ip); 321 return -EIO; 322 } 323 324 error = gfs2_meta_inode_buffer(ip, &dibh); 325 if (error) 326 return error; 327 328 if (diff > 0) 329 inc_nlink(&ip->i_inode); 330 else 331 drop_nlink(&ip->i_inode); 332 333 ip->i_inode.i_ctime.tv_sec = get_seconds(); 334 335 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 336 gfs2_dinode_out(ip, dibh->b_data); 337 brelse(dibh); 338 mark_inode_dirty(&ip->i_inode); 339 340 if (ip->i_inode.i_nlink == 0) { 341 struct gfs2_rgrpd *rgd; 342 struct gfs2_holder ri_gh, rg_gh; 343 344 error = gfs2_rindex_hold(sdp, &ri_gh); 345 if (error) 346 goto out; 347 error = -EIO; 348 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 349 if (!rgd) 350 goto out_norgrp; 351 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh); 352 if (error) 353 goto out_norgrp; 354 355 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 356 gfs2_glock_dq_uninit(&rg_gh); 357 out_norgrp: 358 gfs2_glock_dq_uninit(&ri_gh); 359 } 360 out: 361 return error; 362 } 363 364 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 365 { 366 struct qstr qstr; 367 gfs2_str2qstr(&qstr, name); 368 return gfs2_lookupi(dip, &qstr, 1, NULL); 369 } 370 371 372 /** 373 * gfs2_lookupi - Look up a filename in a directory and return its inode 374 * @d_gh: An initialized holder for the directory glock 375 * @name: The name of the inode to look for 376 * @is_root: If 1, ignore the caller's permissions 377 * @i_gh: An uninitialized holder for the new inode glock 378 * 379 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 380 * @is_root is true. 381 * 382 * Returns: errno 383 */ 384 385 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 386 int is_root, struct nameidata *nd) 387 { 388 struct super_block *sb = dir->i_sb; 389 struct gfs2_inode *dip = GFS2_I(dir); 390 struct gfs2_holder d_gh; 391 struct gfs2_inum_host inum; 392 unsigned int type; 393 int error = 0; 394 struct inode *inode = NULL; 395 396 if (!name->len || name->len > GFS2_FNAMESIZE) 397 return ERR_PTR(-ENAMETOOLONG); 398 399 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 400 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 401 dir == sb->s_root->d_inode)) { 402 igrab(dir); 403 return dir; 404 } 405 406 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 407 if (error) 408 return ERR_PTR(error); 409 410 if (!is_root) { 411 error = permission(dir, MAY_EXEC, NULL); 412 if (error) 413 goto out; 414 } 415 416 error = gfs2_dir_search(dir, name, &inum, &type); 417 if (error) 418 goto out; 419 420 inode = gfs2_inode_lookup(sb, &inum, type); 421 422 out: 423 gfs2_glock_dq_uninit(&d_gh); 424 if (error == -ENOENT) 425 return NULL; 426 return inode; 427 } 428 429 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino) 430 { 431 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 432 struct buffer_head *bh; 433 struct gfs2_inum_range_host ir; 434 int error; 435 436 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 437 if (error) 438 return error; 439 mutex_lock(&sdp->sd_inum_mutex); 440 441 error = gfs2_meta_inode_buffer(ip, &bh); 442 if (error) { 443 mutex_unlock(&sdp->sd_inum_mutex); 444 gfs2_trans_end(sdp); 445 return error; 446 } 447 448 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 449 450 if (ir.ir_length) { 451 *formal_ino = ir.ir_start++; 452 ir.ir_length--; 453 gfs2_trans_add_bh(ip->i_gl, bh, 1); 454 gfs2_inum_range_out(&ir, 455 bh->b_data + sizeof(struct gfs2_dinode)); 456 brelse(bh); 457 mutex_unlock(&sdp->sd_inum_mutex); 458 gfs2_trans_end(sdp); 459 return 0; 460 } 461 462 brelse(bh); 463 464 mutex_unlock(&sdp->sd_inum_mutex); 465 gfs2_trans_end(sdp); 466 467 return 1; 468 } 469 470 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino) 471 { 472 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 473 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 474 struct gfs2_holder gh; 475 struct buffer_head *bh; 476 struct gfs2_inum_range_host ir; 477 int error; 478 479 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 480 if (error) 481 return error; 482 483 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 484 if (error) 485 goto out; 486 mutex_lock(&sdp->sd_inum_mutex); 487 488 error = gfs2_meta_inode_buffer(ip, &bh); 489 if (error) 490 goto out_end_trans; 491 492 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 493 494 if (!ir.ir_length) { 495 struct buffer_head *m_bh; 496 u64 x, y; 497 __be64 z; 498 499 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 500 if (error) 501 goto out_brelse; 502 503 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 504 x = y = be64_to_cpu(z); 505 ir.ir_start = x; 506 ir.ir_length = GFS2_INUM_QUANTUM; 507 x += GFS2_INUM_QUANTUM; 508 if (x < y) 509 gfs2_consist_inode(m_ip); 510 z = cpu_to_be64(x); 511 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 512 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z; 513 514 brelse(m_bh); 515 } 516 517 *formal_ino = ir.ir_start++; 518 ir.ir_length--; 519 520 gfs2_trans_add_bh(ip->i_gl, bh, 1); 521 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 522 523 out_brelse: 524 brelse(bh); 525 out_end_trans: 526 mutex_unlock(&sdp->sd_inum_mutex); 527 gfs2_trans_end(sdp); 528 out: 529 gfs2_glock_dq_uninit(&gh); 530 return error; 531 } 532 533 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum) 534 { 535 int error; 536 537 error = pick_formal_ino_1(sdp, inum); 538 if (error <= 0) 539 return error; 540 541 error = pick_formal_ino_2(sdp, inum); 542 543 return error; 544 } 545 546 /** 547 * create_ok - OK to create a new on-disk inode here? 548 * @dip: Directory in which dinode is to be created 549 * @name: Name of new dinode 550 * @mode: 551 * 552 * Returns: errno 553 */ 554 555 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 556 unsigned int mode) 557 { 558 int error; 559 560 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 561 if (error) 562 return error; 563 564 /* Don't create entries in an unlinked directory */ 565 if (!dip->i_inode.i_nlink) 566 return -EPERM; 567 568 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL); 569 switch (error) { 570 case -ENOENT: 571 error = 0; 572 break; 573 case 0: 574 return -EEXIST; 575 default: 576 return error; 577 } 578 579 if (dip->i_di.di_entries == (u32)-1) 580 return -EFBIG; 581 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1) 582 return -EMLINK; 583 584 return 0; 585 } 586 587 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 588 unsigned int *uid, unsigned int *gid) 589 { 590 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 591 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) { 592 if (S_ISDIR(*mode)) 593 *mode |= S_ISUID; 594 else if (dip->i_inode.i_uid != current->fsuid) 595 *mode &= ~07111; 596 *uid = dip->i_inode.i_uid; 597 } else 598 *uid = current->fsuid; 599 600 if (dip->i_inode.i_mode & S_ISGID) { 601 if (S_ISDIR(*mode)) 602 *mode |= S_ISGID; 603 *gid = dip->i_inode.i_gid; 604 } else 605 *gid = current->fsgid; 606 } 607 608 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum, 609 u64 *generation) 610 { 611 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 612 int error; 613 614 gfs2_alloc_get(dip); 615 616 dip->i_alloc.al_requested = RES_DINODE; 617 error = gfs2_inplace_reserve(dip); 618 if (error) 619 goto out; 620 621 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 622 if (error) 623 goto out_ipreserv; 624 625 inum->no_addr = gfs2_alloc_di(dip, generation); 626 627 gfs2_trans_end(sdp); 628 629 out_ipreserv: 630 gfs2_inplace_release(dip); 631 out: 632 gfs2_alloc_put(dip); 633 return error; 634 } 635 636 /** 637 * init_dinode - Fill in a new dinode structure 638 * @dip: the directory this inode is being created in 639 * @gl: The glock covering the new inode 640 * @inum: the inode number 641 * @mode: the file permissions 642 * @uid: 643 * @gid: 644 * 645 */ 646 647 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 648 const struct gfs2_inum_host *inum, unsigned int mode, 649 unsigned int uid, unsigned int gid, 650 const u64 *generation, dev_t dev) 651 { 652 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 653 struct gfs2_dinode *di; 654 struct buffer_head *dibh; 655 656 dibh = gfs2_meta_new(gl, inum->no_addr); 657 gfs2_trans_add_bh(gl, dibh, 1); 658 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 659 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 660 di = (struct gfs2_dinode *)dibh->b_data; 661 662 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 663 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 664 di->di_mode = cpu_to_be32(mode); 665 di->di_uid = cpu_to_be32(uid); 666 di->di_gid = cpu_to_be32(gid); 667 di->di_nlink = 0; 668 di->di_size = 0; 669 di->di_blocks = cpu_to_be64(1); 670 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 671 di->di_major = cpu_to_be32(MAJOR(dev)); 672 di->di_minor = cpu_to_be32(MINOR(dev)); 673 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 674 di->di_generation = cpu_to_be64(*generation); 675 di->di_flags = 0; 676 677 if (S_ISREG(mode)) { 678 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 679 gfs2_tune_get(sdp, gt_new_files_jdata)) 680 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 681 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 682 gfs2_tune_get(sdp, gt_new_files_directio)) 683 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 684 } else if (S_ISDIR(mode)) { 685 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 686 GFS2_DIF_INHERIT_DIRECTIO); 687 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 688 GFS2_DIF_INHERIT_JDATA); 689 } 690 691 di->__pad1 = 0; 692 di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0); 693 di->di_height = 0; 694 di->__pad2 = 0; 695 di->__pad3 = 0; 696 di->di_depth = 0; 697 di->di_entries = 0; 698 memset(&di->__pad4, 0, sizeof(di->__pad4)); 699 di->di_eattr = 0; 700 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 701 702 brelse(dibh); 703 } 704 705 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 706 unsigned int mode, const struct gfs2_inum_host *inum, 707 const u64 *generation, dev_t dev) 708 { 709 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 710 unsigned int uid, gid; 711 int error; 712 713 munge_mode_uid_gid(dip, &mode, &uid, &gid); 714 gfs2_alloc_get(dip); 715 716 error = gfs2_quota_lock(dip, uid, gid); 717 if (error) 718 goto out; 719 720 error = gfs2_quota_check(dip, uid, gid); 721 if (error) 722 goto out_quota; 723 724 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0); 725 if (error) 726 goto out_quota; 727 728 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev); 729 gfs2_quota_change(dip, +1, uid, gid); 730 gfs2_trans_end(sdp); 731 732 out_quota: 733 gfs2_quota_unlock(dip); 734 out: 735 gfs2_alloc_put(dip); 736 return error; 737 } 738 739 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name, 740 struct gfs2_inode *ip) 741 { 742 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 743 struct gfs2_alloc *al; 744 int alloc_required; 745 struct buffer_head *dibh; 746 int error; 747 748 al = gfs2_alloc_get(dip); 749 750 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 751 if (error) 752 goto fail; 753 754 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name); 755 if (alloc_required < 0) 756 goto fail; 757 if (alloc_required) { 758 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid); 759 if (error) 760 goto fail_quota_locks; 761 762 al->al_requested = sdp->sd_max_dirres; 763 764 error = gfs2_inplace_reserve(dip); 765 if (error) 766 goto fail_quota_locks; 767 768 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 769 al->al_rgd->rd_ri.ri_length + 770 2 * RES_DINODE + 771 RES_STATFS + RES_QUOTA, 0); 772 if (error) 773 goto fail_ipreserv; 774 } else { 775 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 776 if (error) 777 goto fail_quota_locks; 778 } 779 780 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_inode.i_mode)); 781 if (error) 782 goto fail_end_trans; 783 784 error = gfs2_meta_inode_buffer(ip, &dibh); 785 if (error) 786 goto fail_end_trans; 787 ip->i_inode.i_nlink = 1; 788 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 789 gfs2_dinode_out(ip, dibh->b_data); 790 brelse(dibh); 791 return 0; 792 793 fail_end_trans: 794 gfs2_trans_end(sdp); 795 796 fail_ipreserv: 797 if (dip->i_alloc.al_rgd) 798 gfs2_inplace_release(dip); 799 800 fail_quota_locks: 801 gfs2_quota_unlock(dip); 802 803 fail: 804 gfs2_alloc_put(dip); 805 return error; 806 } 807 808 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip) 809 { 810 int err; 811 size_t len; 812 void *value; 813 char *name; 814 struct gfs2_ea_request er; 815 816 err = security_inode_init_security(&ip->i_inode, &dip->i_inode, 817 &name, &value, &len); 818 819 if (err) { 820 if (err == -EOPNOTSUPP) 821 return 0; 822 return err; 823 } 824 825 memset(&er, 0, sizeof(struct gfs2_ea_request)); 826 827 er.er_type = GFS2_EATYPE_SECURITY; 828 er.er_name = name; 829 er.er_data = value; 830 er.er_name_len = strlen(name); 831 er.er_data_len = len; 832 833 err = gfs2_ea_set_i(ip, &er); 834 835 kfree(value); 836 kfree(name); 837 838 return err; 839 } 840 841 /** 842 * gfs2_createi - Create a new inode 843 * @ghs: An array of two holders 844 * @name: The name of the new file 845 * @mode: the permissions on the new inode 846 * 847 * @ghs[0] is an initialized holder for the directory 848 * @ghs[1] is the holder for the inode lock 849 * 850 * If the return value is not NULL, the glocks on both the directory and the new 851 * file are held. A transaction has been started and an inplace reservation 852 * is held, as well. 853 * 854 * Returns: An inode 855 */ 856 857 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 858 unsigned int mode, dev_t dev) 859 { 860 struct inode *inode; 861 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 862 struct inode *dir = &dip->i_inode; 863 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 864 struct gfs2_inum_host inum; 865 int error; 866 u64 generation; 867 868 if (!name->len || name->len > GFS2_FNAMESIZE) 869 return ERR_PTR(-ENAMETOOLONG); 870 871 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 872 error = gfs2_glock_nq(ghs); 873 if (error) 874 goto fail; 875 876 error = create_ok(dip, name, mode); 877 if (error) 878 goto fail_gunlock; 879 880 error = pick_formal_ino(sdp, &inum.no_formal_ino); 881 if (error) 882 goto fail_gunlock; 883 884 error = alloc_dinode(dip, &inum, &generation); 885 if (error) 886 goto fail_gunlock; 887 888 if (inum.no_addr < dip->i_num.no_addr) { 889 gfs2_glock_dq(ghs); 890 891 error = gfs2_glock_nq_num(sdp, inum.no_addr, 892 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 893 GL_SKIP, ghs + 1); 894 if (error) { 895 return ERR_PTR(error); 896 } 897 898 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 899 error = gfs2_glock_nq(ghs); 900 if (error) { 901 gfs2_glock_dq_uninit(ghs + 1); 902 return ERR_PTR(error); 903 } 904 905 error = create_ok(dip, name, mode); 906 if (error) 907 goto fail_gunlock2; 908 } else { 909 error = gfs2_glock_nq_num(sdp, inum.no_addr, 910 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 911 GL_SKIP, ghs + 1); 912 if (error) 913 goto fail_gunlock; 914 } 915 916 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev); 917 if (error) 918 goto fail_gunlock2; 919 920 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode)); 921 if (IS_ERR(inode)) 922 goto fail_gunlock2; 923 924 error = gfs2_inode_refresh(GFS2_I(inode)); 925 if (error) 926 goto fail_iput; 927 928 error = gfs2_acl_create(dip, GFS2_I(inode)); 929 if (error) 930 goto fail_iput; 931 932 error = gfs2_security_init(dip, GFS2_I(inode)); 933 if (error) 934 goto fail_iput; 935 936 error = link_dinode(dip, name, GFS2_I(inode)); 937 if (error) 938 goto fail_iput; 939 940 if (!inode) 941 return ERR_PTR(-ENOMEM); 942 return inode; 943 944 fail_iput: 945 iput(inode); 946 fail_gunlock2: 947 gfs2_glock_dq_uninit(ghs + 1); 948 fail_gunlock: 949 gfs2_glock_dq(ghs); 950 fail: 951 return ERR_PTR(error); 952 } 953 954 /** 955 * gfs2_rmdiri - Remove a directory 956 * @dip: The parent directory of the directory to be removed 957 * @name: The name of the directory to be removed 958 * @ip: The GFS2 inode of the directory to be removed 959 * 960 * Assumes Glocks on dip and ip are held 961 * 962 * Returns: errno 963 */ 964 965 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 966 struct gfs2_inode *ip) 967 { 968 struct qstr dotname; 969 int error; 970 971 if (ip->i_di.di_entries != 2) { 972 if (gfs2_consist_inode(ip)) 973 gfs2_dinode_print(ip); 974 return -EIO; 975 } 976 977 error = gfs2_dir_del(dip, name); 978 if (error) 979 return error; 980 981 error = gfs2_change_nlink(dip, -1); 982 if (error) 983 return error; 984 985 gfs2_str2qstr(&dotname, "."); 986 error = gfs2_dir_del(ip, &dotname); 987 if (error) 988 return error; 989 990 gfs2_str2qstr(&dotname, ".."); 991 error = gfs2_dir_del(ip, &dotname); 992 if (error) 993 return error; 994 995 /* It looks odd, but it really should be done twice */ 996 error = gfs2_change_nlink(ip, -1); 997 if (error) 998 return error; 999 1000 error = gfs2_change_nlink(ip, -1); 1001 if (error) 1002 return error; 1003 1004 return error; 1005 } 1006 1007 /* 1008 * gfs2_unlink_ok - check to see that a inode is still in a directory 1009 * @dip: the directory 1010 * @name: the name of the file 1011 * @ip: the inode 1012 * 1013 * Assumes that the lock on (at least) @dip is held. 1014 * 1015 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 1016 */ 1017 1018 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 1019 struct gfs2_inode *ip) 1020 { 1021 struct gfs2_inum_host inum; 1022 unsigned int type; 1023 int error; 1024 1025 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 1026 return -EPERM; 1027 1028 if ((dip->i_inode.i_mode & S_ISVTX) && 1029 dip->i_inode.i_uid != current->fsuid && 1030 ip->i_inode.i_uid != current->fsuid && !capable(CAP_FOWNER)) 1031 return -EPERM; 1032 1033 if (IS_APPEND(&dip->i_inode)) 1034 return -EPERM; 1035 1036 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 1037 if (error) 1038 return error; 1039 1040 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type); 1041 if (error) 1042 return error; 1043 1044 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1045 return -ENOENT; 1046 1047 if (IF2DT(ip->i_inode.i_mode) != type) { 1048 gfs2_consist_inode(dip); 1049 return -EIO; 1050 } 1051 1052 return 0; 1053 } 1054 1055 /* 1056 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1057 * @this: move this 1058 * @to: to here 1059 * 1060 * Follow @to back to the root and make sure we don't encounter @this 1061 * Assumes we already hold the rename lock. 1062 * 1063 * Returns: errno 1064 */ 1065 1066 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1067 { 1068 struct inode *dir = &to->i_inode; 1069 struct super_block *sb = dir->i_sb; 1070 struct inode *tmp; 1071 struct qstr dotdot; 1072 int error = 0; 1073 1074 gfs2_str2qstr(&dotdot, ".."); 1075 1076 igrab(dir); 1077 1078 for (;;) { 1079 if (dir == &this->i_inode) { 1080 error = -EINVAL; 1081 break; 1082 } 1083 if (dir == sb->s_root->d_inode) { 1084 error = 0; 1085 break; 1086 } 1087 1088 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1089 if (IS_ERR(tmp)) { 1090 error = PTR_ERR(tmp); 1091 break; 1092 } 1093 1094 iput(dir); 1095 dir = tmp; 1096 } 1097 1098 iput(dir); 1099 1100 return error; 1101 } 1102 1103 /** 1104 * gfs2_readlinki - return the contents of a symlink 1105 * @ip: the symlink's inode 1106 * @buf: a pointer to the buffer to be filled 1107 * @len: a pointer to the length of @buf 1108 * 1109 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1110 * to be freed by the caller. 1111 * 1112 * Returns: errno 1113 */ 1114 1115 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1116 { 1117 struct gfs2_holder i_gh; 1118 struct buffer_head *dibh; 1119 unsigned int x; 1120 int error; 1121 1122 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1123 error = gfs2_glock_nq_atime(&i_gh); 1124 if (error) { 1125 gfs2_holder_uninit(&i_gh); 1126 return error; 1127 } 1128 1129 if (!ip->i_di.di_size) { 1130 gfs2_consist_inode(ip); 1131 error = -EIO; 1132 goto out; 1133 } 1134 1135 error = gfs2_meta_inode_buffer(ip, &dibh); 1136 if (error) 1137 goto out; 1138 1139 x = ip->i_di.di_size + 1; 1140 if (x > *len) { 1141 *buf = kmalloc(x, GFP_KERNEL); 1142 if (!*buf) { 1143 error = -ENOMEM; 1144 goto out_brelse; 1145 } 1146 } 1147 1148 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1149 *len = x; 1150 1151 out_brelse: 1152 brelse(dibh); 1153 out: 1154 gfs2_glock_dq_uninit(&i_gh); 1155 return error; 1156 } 1157 1158 /** 1159 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1160 * conditionally update the inode's atime 1161 * @gh: the holder to acquire 1162 * 1163 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1164 * Update if the difference between the current time and the inode's current 1165 * atime is greater than an interval specified at mount. 1166 * 1167 * Returns: errno 1168 */ 1169 1170 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1171 { 1172 struct gfs2_glock *gl = gh->gh_gl; 1173 struct gfs2_sbd *sdp = gl->gl_sbd; 1174 struct gfs2_inode *ip = gl->gl_object; 1175 s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1176 unsigned int state; 1177 int flags; 1178 int error; 1179 1180 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1181 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1182 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1183 return -EINVAL; 1184 1185 state = gh->gh_state; 1186 flags = gh->gh_flags; 1187 1188 error = gfs2_glock_nq(gh); 1189 if (error) 1190 return error; 1191 1192 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1193 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1194 return 0; 1195 1196 curtime = get_seconds(); 1197 if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) { 1198 gfs2_glock_dq(gh); 1199 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1200 gh); 1201 error = gfs2_glock_nq(gh); 1202 if (error) 1203 return error; 1204 1205 /* Verify that atime hasn't been updated while we were 1206 trying to get exclusive lock. */ 1207 1208 curtime = get_seconds(); 1209 if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) { 1210 struct buffer_head *dibh; 1211 struct gfs2_dinode *di; 1212 1213 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1214 if (error == -EROFS) 1215 return 0; 1216 if (error) 1217 goto fail; 1218 1219 error = gfs2_meta_inode_buffer(ip, &dibh); 1220 if (error) 1221 goto fail_end_trans; 1222 1223 ip->i_inode.i_atime.tv_sec = curtime; 1224 1225 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1226 di = (struct gfs2_dinode *)dibh->b_data; 1227 di->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec); 1228 brelse(dibh); 1229 1230 gfs2_trans_end(sdp); 1231 } 1232 1233 /* If someone else has asked for the glock, 1234 unlock and let them have it. Then reacquire 1235 in the original state. */ 1236 if (gfs2_glock_is_blocking(gl)) { 1237 gfs2_glock_dq(gh); 1238 gfs2_holder_reinit(state, flags, gh); 1239 return gfs2_glock_nq(gh); 1240 } 1241 } 1242 1243 return 0; 1244 1245 fail_end_trans: 1246 gfs2_trans_end(sdp); 1247 fail: 1248 gfs2_glock_dq(gh); 1249 return error; 1250 } 1251 1252 /** 1253 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1254 * @arg_a: the first structure 1255 * @arg_b: the second structure 1256 * 1257 * Returns: 1 if A > B 1258 * -1 if A < B 1259 * 0 if A == B 1260 */ 1261 1262 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1263 { 1264 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; 1265 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; 1266 const struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1267 const struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1268 1269 if (a->ln_number > b->ln_number) 1270 return 1; 1271 if (a->ln_number < b->ln_number) 1272 return -1; 1273 if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE) 1274 return 1; 1275 if (gh_a->gh_state == LM_ST_SHARED && (gh_b->gh_flags & GL_ATIME)) 1276 return 1; 1277 1278 return 0; 1279 } 1280 1281 /** 1282 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1283 * atime update 1284 * @num_gh: the number of structures 1285 * @ghs: an array of struct gfs2_holder structures 1286 * 1287 * Returns: 0 on success (all glocks acquired), 1288 * errno on failure (no glocks acquired) 1289 */ 1290 1291 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1292 { 1293 struct gfs2_holder **p; 1294 unsigned int x; 1295 int error = 0; 1296 1297 if (!num_gh) 1298 return 0; 1299 1300 if (num_gh == 1) { 1301 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1302 if (ghs->gh_flags & GL_ATIME) 1303 error = gfs2_glock_nq_atime(ghs); 1304 else 1305 error = gfs2_glock_nq(ghs); 1306 return error; 1307 } 1308 1309 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1310 if (!p) 1311 return -ENOMEM; 1312 1313 for (x = 0; x < num_gh; x++) 1314 p[x] = &ghs[x]; 1315 1316 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1317 1318 for (x = 0; x < num_gh; x++) { 1319 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1320 1321 if (p[x]->gh_flags & GL_ATIME) 1322 error = gfs2_glock_nq_atime(p[x]); 1323 else 1324 error = gfs2_glock_nq(p[x]); 1325 1326 if (error) { 1327 while (x--) 1328 gfs2_glock_dq(p[x]); 1329 break; 1330 } 1331 } 1332 1333 kfree(p); 1334 return error; 1335 } 1336 1337 1338 static int 1339 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1340 { 1341 struct buffer_head *dibh; 1342 int error; 1343 1344 error = gfs2_meta_inode_buffer(ip, &dibh); 1345 if (!error) { 1346 error = inode_setattr(&ip->i_inode, attr); 1347 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1348 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1349 gfs2_dinode_out(ip, dibh->b_data); 1350 brelse(dibh); 1351 } 1352 return error; 1353 } 1354 1355 /** 1356 * gfs2_setattr_simple - 1357 * @ip: 1358 * @attr: 1359 * 1360 * Called with a reference on the vnode. 1361 * 1362 * Returns: errno 1363 */ 1364 1365 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1366 { 1367 int error; 1368 1369 if (current->journal_info) 1370 return __gfs2_setattr_simple(ip, attr); 1371 1372 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1373 if (error) 1374 return error; 1375 1376 error = __gfs2_setattr_simple(ip, attr); 1377 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1378 return error; 1379 } 1380 1381