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 v.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 20 #include "gfs2.h" 21 #include "lm_interface.h" 22 #include "incore.h" 23 #include "acl.h" 24 #include "bmap.h" 25 #include "dir.h" 26 #include "eattr.h" 27 #include "glock.h" 28 #include "glops.h" 29 #include "inode.h" 30 #include "log.h" 31 #include "meta_io.h" 32 #include "ops_address.h" 33 #include "ops_file.h" 34 #include "ops_inode.h" 35 #include "quota.h" 36 #include "rgrp.h" 37 #include "trans.h" 38 #include "util.h" 39 40 /** 41 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode 42 * @ip: The GFS2 inode (with embedded disk inode data) 43 * @inode: The Linux VFS inode 44 * 45 */ 46 47 void gfs2_inode_attr_in(struct gfs2_inode *ip) 48 { 49 struct inode *inode = &ip->i_inode; 50 51 inode->i_ino = ip->i_num.no_addr; 52 53 switch (ip->i_di.di_mode & S_IFMT) { 54 case S_IFBLK: 55 case S_IFCHR: 56 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor); 57 break; 58 default: 59 inode->i_rdev = 0; 60 break; 61 }; 62 63 inode->i_mode = ip->i_di.di_mode; 64 inode->i_nlink = ip->i_di.di_nlink; 65 inode->i_uid = ip->i_di.di_uid; 66 inode->i_gid = ip->i_di.di_gid; 67 i_size_write(inode, ip->i_di.di_size); 68 inode->i_atime.tv_sec = ip->i_di.di_atime; 69 inode->i_mtime.tv_sec = ip->i_di.di_mtime; 70 inode->i_ctime.tv_sec = ip->i_di.di_ctime; 71 inode->i_atime.tv_nsec = 0; 72 inode->i_mtime.tv_nsec = 0; 73 inode->i_ctime.tv_nsec = 0; 74 inode->i_blksize = PAGE_SIZE; 75 inode->i_blocks = ip->i_di.di_blocks << 76 (GFS2_SB(inode)->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT); 77 78 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE) 79 inode->i_flags |= S_IMMUTABLE; 80 else 81 inode->i_flags &= ~S_IMMUTABLE; 82 83 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY) 84 inode->i_flags |= S_APPEND; 85 else 86 inode->i_flags &= ~S_APPEND; 87 } 88 89 /** 90 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode 91 * @ip: The GFS2 inode 92 * 93 * Only copy out the attributes that we want the VFS layer 94 * to be able to modify. 95 */ 96 97 void gfs2_inode_attr_out(struct gfs2_inode *ip) 98 { 99 struct inode *inode = &ip->i_inode; 100 101 gfs2_assert_withdraw(GFS2_SB(inode), 102 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT)); 103 ip->i_di.di_mode = inode->i_mode; 104 ip->i_di.di_uid = inode->i_uid; 105 ip->i_di.di_gid = inode->i_gid; 106 ip->i_di.di_atime = inode->i_atime.tv_sec; 107 ip->i_di.di_mtime = inode->i_mtime.tv_sec; 108 ip->i_di.di_ctime = inode->i_ctime.tv_sec; 109 } 110 111 static int iget_test(struct inode *inode, void *opaque) 112 { 113 struct gfs2_inode *ip = GFS2_I(inode); 114 struct gfs2_inum *inum = opaque; 115 116 if (ip && ip->i_num.no_addr == inum->no_addr) 117 return 1; 118 119 return 0; 120 } 121 122 static int iget_set(struct inode *inode, void *opaque) 123 { 124 struct gfs2_inode *ip = GFS2_I(inode); 125 struct gfs2_inum *inum = opaque; 126 127 ip->i_num = *inum; 128 return 0; 129 } 130 131 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum *inum) 132 { 133 return ilookup5(sb, (unsigned long)inum->no_formal_ino, 134 iget_test, inum); 135 } 136 137 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum) 138 { 139 return iget5_locked(sb, (unsigned long)inum->no_formal_ino, 140 iget_test, iget_set, inum); 141 } 142 143 /** 144 * gfs2_inode_lookup - Lookup an inode 145 * @sb: The super block 146 * @inum: The inode number 147 * @type: The type of the inode 148 * 149 * Returns: A VFS inode, or an error 150 */ 151 152 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum *inum, unsigned int type) 153 { 154 struct inode *inode = gfs2_iget(sb, inum); 155 struct gfs2_inode *ip = GFS2_I(inode); 156 struct gfs2_glock *io_gl; 157 int error; 158 159 if (inode->i_state & I_NEW) { 160 struct gfs2_sbd *sdp = GFS2_SB(inode); 161 umode_t mode = DT2IF(type); 162 inode->u.generic_ip = ip; 163 inode->i_mode = mode; 164 165 if (S_ISREG(mode)) { 166 inode->i_op = &gfs2_file_iops; 167 inode->i_fop = &gfs2_file_fops; 168 inode->i_mapping->a_ops = &gfs2_file_aops; 169 } else if (S_ISDIR(mode)) { 170 inode->i_op = &gfs2_dir_iops; 171 inode->i_fop = &gfs2_dir_fops; 172 } else if (S_ISLNK(mode)) { 173 inode->i_op = &gfs2_symlink_iops; 174 } else { 175 inode->i_op = &gfs2_dev_iops; 176 } 177 178 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl); 179 if (unlikely(error)) 180 goto fail; 181 ip->i_gl->gl_object = ip; 182 183 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl); 184 if (unlikely(error)) 185 goto fail_put; 186 187 ip->i_vn = ip->i_gl->gl_vn - 1; 188 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh); 189 if (unlikely(error)) 190 goto fail_iopen; 191 192 gfs2_glock_put(io_gl); 193 unlock_new_inode(inode); 194 } 195 196 return inode; 197 fail_iopen: 198 gfs2_glock_put(io_gl); 199 fail_put: 200 ip->i_gl->gl_object = NULL; 201 gfs2_glock_put(ip->i_gl); 202 fail: 203 iput(inode); 204 return ERR_PTR(error); 205 } 206 207 /** 208 * gfs2_inode_refresh - Refresh the incore copy of the dinode 209 * @ip: The GFS2 inode 210 * 211 * Returns: errno 212 */ 213 214 int gfs2_inode_refresh(struct gfs2_inode *ip) 215 { 216 struct buffer_head *dibh; 217 int error; 218 219 error = gfs2_meta_inode_buffer(ip, &dibh); 220 if (error) 221 return error; 222 223 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) { 224 brelse(dibh); 225 return -EIO; 226 } 227 228 gfs2_dinode_in(&ip->i_di, dibh->b_data); 229 230 brelse(dibh); 231 232 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) { 233 if (gfs2_consist_inode(ip)) 234 gfs2_dinode_print(&ip->i_di); 235 return -EIO; 236 } 237 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino) 238 return -ESTALE; 239 240 ip->i_vn = ip->i_gl->gl_vn; 241 242 return 0; 243 } 244 245 int gfs2_dinode_dealloc(struct gfs2_inode *ip) 246 { 247 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 248 struct gfs2_alloc *al; 249 struct gfs2_rgrpd *rgd; 250 int error; 251 252 if (ip->i_di.di_blocks != 1) { 253 if (gfs2_consist_inode(ip)) 254 gfs2_dinode_print(&ip->i_di); 255 return -EIO; 256 } 257 258 al = gfs2_alloc_get(ip); 259 260 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 261 if (error) 262 goto out; 263 264 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 265 if (error) 266 goto out_qs; 267 268 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 269 if (!rgd) { 270 gfs2_consist_inode(ip); 271 error = -EIO; 272 goto out_rindex_relse; 273 } 274 275 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 276 &al->al_rgd_gh); 277 if (error) 278 goto out_rindex_relse; 279 280 error = gfs2_trans_begin(sdp, RES_RG_BIT + 281 RES_STATFS + RES_QUOTA, 1); 282 if (error) 283 goto out_rg_gunlock; 284 285 gfs2_trans_add_gl(ip->i_gl); 286 287 gfs2_free_di(rgd, ip); 288 289 gfs2_trans_end(sdp); 290 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 291 292 out_rg_gunlock: 293 gfs2_glock_dq_uninit(&al->al_rgd_gh); 294 out_rindex_relse: 295 gfs2_glock_dq_uninit(&al->al_ri_gh); 296 out_qs: 297 gfs2_quota_unhold(ip); 298 out: 299 gfs2_alloc_put(ip); 300 return error; 301 } 302 303 /** 304 * gfs2_change_nlink - Change nlink count on inode 305 * @ip: The GFS2 inode 306 * @diff: The change in the nlink count required 307 * 308 * Returns: errno 309 */ 310 311 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 312 { 313 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info; 314 struct buffer_head *dibh; 315 uint32_t nlink; 316 int error; 317 318 BUG_ON(ip->i_di.di_nlink != ip->i_inode.i_nlink); 319 nlink = ip->i_di.di_nlink + diff; 320 321 /* If we are reducing the nlink count, but the new value ends up being 322 bigger than the old one, we must have underflowed. */ 323 if (diff < 0 && nlink > ip->i_di.di_nlink) { 324 if (gfs2_consist_inode(ip)) 325 gfs2_dinode_print(&ip->i_di); 326 return -EIO; 327 } 328 329 error = gfs2_meta_inode_buffer(ip, &dibh); 330 if (error) 331 return error; 332 333 ip->i_di.di_nlink = nlink; 334 ip->i_di.di_ctime = get_seconds(); 335 ip->i_inode.i_nlink = nlink; 336 337 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 338 gfs2_dinode_out(&ip->i_di, dibh->b_data); 339 brelse(dibh); 340 mark_inode_dirty(&ip->i_inode); 341 342 if (ip->i_di.di_nlink == 0) { 343 struct gfs2_rgrpd *rgd; 344 struct gfs2_holder ri_gh, rg_gh; 345 346 error = gfs2_rindex_hold(sdp, &ri_gh); 347 if (error) 348 goto out; 349 error = -EIO; 350 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 351 if (!rgd) 352 goto out_norgrp; 353 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh); 354 if (error) 355 goto out_norgrp; 356 357 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 358 gfs2_glock_dq_uninit(&rg_gh); 359 out_norgrp: 360 gfs2_glock_dq_uninit(&ri_gh); 361 } 362 out: 363 return error; 364 } 365 366 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 367 { 368 struct qstr qstr; 369 gfs2_str2qstr(&qstr, name); 370 return gfs2_lookupi(dip, &qstr, 1, NULL); 371 } 372 373 374 /** 375 * gfs2_lookupi - Look up a filename in a directory and return its inode 376 * @d_gh: An initialized holder for the directory glock 377 * @name: The name of the inode to look for 378 * @is_root: If 1, ignore the caller's permissions 379 * @i_gh: An uninitialized holder for the new inode glock 380 * 381 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 382 * @is_root is true. 383 * 384 * Returns: errno 385 */ 386 387 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 388 int is_root, struct nameidata *nd) 389 390 { 391 struct super_block *sb = dir->i_sb; 392 struct gfs2_inode *dip = GFS2_I(dir); 393 struct gfs2_holder d_gh; 394 struct gfs2_inum inum; 395 unsigned int type; 396 int error = 0; 397 struct inode *inode = NULL; 398 399 if (!name->len || name->len > GFS2_FNAMESIZE) 400 return ERR_PTR(-ENAMETOOLONG); 401 402 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 403 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 404 dir == sb->s_root->d_inode)) { 405 igrab(dir); 406 return dir; 407 } 408 409 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 410 if (error) 411 return ERR_PTR(error); 412 413 if (!is_root) { 414 error = permission(dir, MAY_EXEC, NULL); 415 if (error) 416 goto out; 417 } 418 419 error = gfs2_dir_search(dir, name, &inum, &type); 420 if (error) 421 goto out; 422 423 inode = gfs2_inode_lookup(sb, &inum, type); 424 425 out: 426 gfs2_glock_dq_uninit(&d_gh); 427 if (error == -ENOENT) 428 return NULL; 429 return inode; 430 } 431 432 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino) 433 { 434 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 435 struct buffer_head *bh; 436 struct gfs2_inum_range ir; 437 int error; 438 439 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 440 if (error) 441 return error; 442 mutex_lock(&sdp->sd_inum_mutex); 443 444 error = gfs2_meta_inode_buffer(ip, &bh); 445 if (error) { 446 mutex_unlock(&sdp->sd_inum_mutex); 447 gfs2_trans_end(sdp); 448 return error; 449 } 450 451 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 452 453 if (ir.ir_length) { 454 *formal_ino = ir.ir_start++; 455 ir.ir_length--; 456 gfs2_trans_add_bh(ip->i_gl, bh, 1); 457 gfs2_inum_range_out(&ir, 458 bh->b_data + sizeof(struct gfs2_dinode)); 459 brelse(bh); 460 mutex_unlock(&sdp->sd_inum_mutex); 461 gfs2_trans_end(sdp); 462 return 0; 463 } 464 465 brelse(bh); 466 467 mutex_unlock(&sdp->sd_inum_mutex); 468 gfs2_trans_end(sdp); 469 470 return 1; 471 } 472 473 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino) 474 { 475 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 476 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 477 struct gfs2_holder gh; 478 struct buffer_head *bh; 479 struct gfs2_inum_range ir; 480 int error; 481 482 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 483 if (error) 484 return error; 485 486 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 487 if (error) 488 goto out; 489 mutex_lock(&sdp->sd_inum_mutex); 490 491 error = gfs2_meta_inode_buffer(ip, &bh); 492 if (error) 493 goto out_end_trans; 494 495 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 496 497 if (!ir.ir_length) { 498 struct buffer_head *m_bh; 499 uint64_t x, y; 500 501 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 502 if (error) 503 goto out_brelse; 504 505 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 506 x = y = be64_to_cpu(x); 507 ir.ir_start = x; 508 ir.ir_length = GFS2_INUM_QUANTUM; 509 x += GFS2_INUM_QUANTUM; 510 if (x < y) 511 gfs2_consist_inode(m_ip); 512 x = cpu_to_be64(x); 513 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 514 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x; 515 516 brelse(m_bh); 517 } 518 519 *formal_ino = ir.ir_start++; 520 ir.ir_length--; 521 522 gfs2_trans_add_bh(ip->i_gl, bh, 1); 523 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 524 525 out_brelse: 526 brelse(bh); 527 528 out_end_trans: 529 mutex_unlock(&sdp->sd_inum_mutex); 530 gfs2_trans_end(sdp); 531 532 out: 533 gfs2_glock_dq_uninit(&gh); 534 535 return error; 536 } 537 538 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum) 539 { 540 int error; 541 542 error = pick_formal_ino_1(sdp, inum); 543 if (error <= 0) 544 return error; 545 546 error = pick_formal_ino_2(sdp, inum); 547 548 return error; 549 } 550 551 /** 552 * create_ok - OK to create a new on-disk inode here? 553 * @dip: Directory in which dinode is to be created 554 * @name: Name of new dinode 555 * @mode: 556 * 557 * Returns: errno 558 */ 559 560 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 561 unsigned int mode) 562 { 563 int error; 564 565 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 566 if (error) 567 return error; 568 569 /* Don't create entries in an unlinked directory */ 570 if (!dip->i_di.di_nlink) 571 return -EPERM; 572 573 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL); 574 switch (error) { 575 case -ENOENT: 576 error = 0; 577 break; 578 case 0: 579 return -EEXIST; 580 default: 581 return error; 582 } 583 584 if (dip->i_di.di_entries == (uint32_t)-1) 585 return -EFBIG; 586 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1) 587 return -EMLINK; 588 589 return 0; 590 } 591 592 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 593 unsigned int *uid, unsigned int *gid) 594 { 595 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 596 (dip->i_di.di_mode & S_ISUID) && 597 dip->i_di.di_uid) { 598 if (S_ISDIR(*mode)) 599 *mode |= S_ISUID; 600 else if (dip->i_di.di_uid != current->fsuid) 601 *mode &= ~07111; 602 *uid = dip->i_di.di_uid; 603 } else 604 *uid = current->fsuid; 605 606 if (dip->i_di.di_mode & S_ISGID) { 607 if (S_ISDIR(*mode)) 608 *mode |= S_ISGID; 609 *gid = dip->i_di.di_gid; 610 } else 611 *gid = current->fsgid; 612 } 613 614 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum *inum, 615 u64 *generation) 616 { 617 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 618 int error; 619 620 gfs2_alloc_get(dip); 621 622 dip->i_alloc.al_requested = RES_DINODE; 623 error = gfs2_inplace_reserve(dip); 624 if (error) 625 goto out; 626 627 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 628 if (error) 629 goto out_ipreserv; 630 631 inum->no_addr = gfs2_alloc_di(dip, generation); 632 633 gfs2_trans_end(sdp); 634 635 out_ipreserv: 636 gfs2_inplace_release(dip); 637 638 out: 639 gfs2_alloc_put(dip); 640 641 return error; 642 } 643 644 /** 645 * init_dinode - Fill in a new dinode structure 646 * @dip: the directory this inode is being created in 647 * @gl: The glock covering the new inode 648 * @inum: the inode number 649 * @mode: the file permissions 650 * @uid: 651 * @gid: 652 * 653 */ 654 655 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 656 const struct gfs2_inum *inum, unsigned int mode, 657 unsigned int uid, unsigned int gid, 658 const u64 *generation) 659 { 660 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 661 struct gfs2_dinode *di; 662 struct buffer_head *dibh; 663 664 dibh = gfs2_meta_new(gl, inum->no_addr); 665 gfs2_trans_add_bh(gl, dibh, 1); 666 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 667 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 668 di = (struct gfs2_dinode *)dibh->b_data; 669 670 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 671 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 672 di->di_mode = cpu_to_be32(mode); 673 di->di_uid = cpu_to_be32(uid); 674 di->di_gid = cpu_to_be32(gid); 675 di->di_nlink = cpu_to_be32(0); 676 di->di_size = cpu_to_be64(0); 677 di->di_blocks = cpu_to_be64(1); 678 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 679 di->di_major = di->di_minor = cpu_to_be32(0); 680 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 681 di->di_generation = cpu_to_be64(*generation); 682 di->di_flags = cpu_to_be32(0); 683 684 if (S_ISREG(mode)) { 685 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 686 gfs2_tune_get(sdp, gt_new_files_jdata)) 687 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 688 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 689 gfs2_tune_get(sdp, gt_new_files_directio)) 690 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 691 } else if (S_ISDIR(mode)) { 692 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 693 GFS2_DIF_INHERIT_DIRECTIO); 694 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 695 GFS2_DIF_INHERIT_JDATA); 696 } 697 698 di->__pad1 = 0; 699 di->di_height = cpu_to_be32(0); 700 di->__pad2 = 0; 701 di->__pad3 = 0; 702 di->di_depth = cpu_to_be16(0); 703 di->di_entries = cpu_to_be32(0); 704 memset(&di->__pad4, 0, sizeof(di->__pad4)); 705 di->di_eattr = cpu_to_be64(0); 706 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 707 708 brelse(dibh); 709 } 710 711 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 712 unsigned int mode, const struct gfs2_inum *inum, 713 const u64 *generation) 714 { 715 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 716 unsigned int uid, gid; 717 int error; 718 719 munge_mode_uid_gid(dip, &mode, &uid, &gid); 720 gfs2_alloc_get(dip); 721 722 error = gfs2_quota_lock(dip, uid, gid); 723 if (error) 724 goto out; 725 726 error = gfs2_quota_check(dip, uid, gid); 727 if (error) 728 goto out_quota; 729 730 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0); 731 if (error) 732 goto out_quota; 733 734 init_dinode(dip, gl, inum, mode, uid, gid, generation); 735 gfs2_quota_change(dip, +1, uid, gid); 736 gfs2_trans_end(sdp); 737 738 out_quota: 739 gfs2_quota_unlock(dip); 740 out: 741 gfs2_alloc_put(dip); 742 return error; 743 } 744 745 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name, 746 struct gfs2_inode *ip) 747 { 748 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 749 struct gfs2_alloc *al; 750 int alloc_required; 751 struct buffer_head *dibh; 752 int error; 753 754 al = gfs2_alloc_get(dip); 755 756 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 757 if (error) 758 goto fail; 759 760 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name); 761 if (alloc_required < 0) 762 goto fail; 763 if (alloc_required) { 764 error = gfs2_quota_check(dip, dip->i_di.di_uid, 765 dip->i_di.di_gid); 766 if (error) 767 goto fail_quota_locks; 768 769 al->al_requested = sdp->sd_max_dirres; 770 771 error = gfs2_inplace_reserve(dip); 772 if (error) 773 goto fail_quota_locks; 774 775 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 776 al->al_rgd->rd_ri.ri_length + 777 2 * RES_DINODE + 778 RES_STATFS + RES_QUOTA, 0); 779 if (error) 780 goto fail_ipreserv; 781 } else { 782 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 783 if (error) 784 goto fail_quota_locks; 785 } 786 787 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_di.di_mode)); 788 if (error) 789 goto fail_end_trans; 790 791 error = gfs2_meta_inode_buffer(ip, &dibh); 792 if (error) 793 goto fail_end_trans; 794 ip->i_di.di_nlink = 1; 795 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 796 gfs2_dinode_out(&ip->i_di, dibh->b_data); 797 brelse(dibh); 798 return 0; 799 800 fail_end_trans: 801 gfs2_trans_end(sdp); 802 803 fail_ipreserv: 804 if (dip->i_alloc.al_rgd) 805 gfs2_inplace_release(dip); 806 807 fail_quota_locks: 808 gfs2_quota_unlock(dip); 809 810 fail: 811 gfs2_alloc_put(dip); 812 return error; 813 } 814 815 /** 816 * gfs2_createi - Create a new inode 817 * @ghs: An array of two holders 818 * @name: The name of the new file 819 * @mode: the permissions on the new inode 820 * 821 * @ghs[0] is an initialized holder for the directory 822 * @ghs[1] is the holder for the inode lock 823 * 824 * If the return value is not NULL, the glocks on both the directory and the new 825 * file are held. A transaction has been started and an inplace reservation 826 * is held, as well. 827 * 828 * Returns: An inode 829 */ 830 831 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 832 unsigned int mode) 833 { 834 struct inode *inode; 835 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 836 struct inode *dir = &dip->i_inode; 837 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 838 struct gfs2_inum inum; 839 int error; 840 u64 generation; 841 842 if (!name->len || name->len > GFS2_FNAMESIZE) 843 return ERR_PTR(-ENAMETOOLONG); 844 845 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 846 error = gfs2_glock_nq(ghs); 847 if (error) 848 goto fail; 849 850 error = create_ok(dip, name, mode); 851 if (error) 852 goto fail_gunlock; 853 854 error = pick_formal_ino(sdp, &inum.no_formal_ino); 855 if (error) 856 goto fail_gunlock; 857 858 error = alloc_dinode(dip, &inum, &generation); 859 if (error) 860 goto fail_gunlock; 861 862 if (inum.no_addr < dip->i_num.no_addr) { 863 gfs2_glock_dq(ghs); 864 865 error = gfs2_glock_nq_num(sdp, inum.no_addr, 866 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 867 GL_SKIP, ghs + 1); 868 if (error) { 869 return ERR_PTR(error); 870 } 871 872 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 873 error = gfs2_glock_nq(ghs); 874 if (error) { 875 gfs2_glock_dq_uninit(ghs + 1); 876 return ERR_PTR(error); 877 } 878 879 error = create_ok(dip, name, mode); 880 if (error) 881 goto fail_gunlock2; 882 } else { 883 error = gfs2_glock_nq_num(sdp, inum.no_addr, 884 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 885 GL_SKIP, ghs + 1); 886 if (error) 887 goto fail_gunlock; 888 } 889 890 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation); 891 if (error) 892 goto fail_gunlock2; 893 894 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode)); 895 if (IS_ERR(inode)) 896 goto fail_gunlock2; 897 898 error = gfs2_inode_refresh(GFS2_I(inode)); 899 if (error) 900 goto fail_iput; 901 902 error = gfs2_acl_create(dip, GFS2_I(inode)); 903 if (error) 904 goto fail_iput; 905 906 error = link_dinode(dip, name, GFS2_I(inode)); 907 if (error) 908 goto fail_iput; 909 910 if (!inode) 911 return ERR_PTR(-ENOMEM); 912 return inode; 913 914 fail_iput: 915 iput(inode); 916 fail_gunlock2: 917 gfs2_glock_dq_uninit(ghs + 1); 918 fail_gunlock: 919 gfs2_glock_dq(ghs); 920 fail: 921 return ERR_PTR(error); 922 } 923 924 /** 925 * gfs2_rmdiri - Remove a directory 926 * @dip: The parent directory of the directory to be removed 927 * @name: The name of the directory to be removed 928 * @ip: The GFS2 inode of the directory to be removed 929 * 930 * Assumes Glocks on dip and ip are held 931 * 932 * Returns: errno 933 */ 934 935 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 936 struct gfs2_inode *ip) 937 { 938 struct qstr dotname; 939 int error; 940 941 if (ip->i_di.di_entries != 2) { 942 if (gfs2_consist_inode(ip)) 943 gfs2_dinode_print(&ip->i_di); 944 return -EIO; 945 } 946 947 error = gfs2_dir_del(dip, name); 948 if (error) 949 return error; 950 951 error = gfs2_change_nlink(dip, -1); 952 if (error) 953 return error; 954 955 gfs2_str2qstr(&dotname, "."); 956 error = gfs2_dir_del(ip, &dotname); 957 if (error) 958 return error; 959 960 gfs2_str2qstr(&dotname, ".."); 961 error = gfs2_dir_del(ip, &dotname); 962 if (error) 963 return error; 964 965 error = gfs2_change_nlink(ip, -2); 966 if (error) 967 return error; 968 969 return error; 970 } 971 972 /* 973 * gfs2_unlink_ok - check to see that a inode is still in a directory 974 * @dip: the directory 975 * @name: the name of the file 976 * @ip: the inode 977 * 978 * Assumes that the lock on (at least) @dip is held. 979 * 980 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 981 */ 982 983 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 984 struct gfs2_inode *ip) 985 { 986 struct gfs2_inum inum; 987 unsigned int type; 988 int error; 989 990 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 991 return -EPERM; 992 993 if ((dip->i_di.di_mode & S_ISVTX) && 994 dip->i_di.di_uid != current->fsuid && 995 ip->i_di.di_uid != current->fsuid && !capable(CAP_FOWNER)) 996 return -EPERM; 997 998 if (IS_APPEND(&dip->i_inode)) 999 return -EPERM; 1000 1001 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 1002 if (error) 1003 return error; 1004 1005 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type); 1006 if (error) 1007 return error; 1008 1009 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1010 return -ENOENT; 1011 1012 if (IF2DT(ip->i_di.di_mode) != type) { 1013 gfs2_consist_inode(dip); 1014 return -EIO; 1015 } 1016 1017 return 0; 1018 } 1019 1020 /* 1021 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1022 * @this: move this 1023 * @to: to here 1024 * 1025 * Follow @to back to the root and make sure we don't encounter @this 1026 * Assumes we already hold the rename lock. 1027 * 1028 * Returns: errno 1029 */ 1030 1031 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1032 { 1033 struct inode *dir = &to->i_inode; 1034 struct super_block *sb = dir->i_sb; 1035 struct inode *tmp; 1036 struct qstr dotdot; 1037 int error = 0; 1038 1039 gfs2_str2qstr(&dotdot, ".."); 1040 1041 igrab(dir); 1042 1043 for (;;) { 1044 if (dir == &this->i_inode) { 1045 error = -EINVAL; 1046 break; 1047 } 1048 if (dir == sb->s_root->d_inode) { 1049 error = 0; 1050 break; 1051 } 1052 1053 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1054 if (IS_ERR(tmp)) { 1055 error = PTR_ERR(tmp); 1056 break; 1057 } 1058 1059 iput(dir); 1060 dir = tmp; 1061 } 1062 1063 iput(dir); 1064 1065 return error; 1066 } 1067 1068 /** 1069 * gfs2_readlinki - return the contents of a symlink 1070 * @ip: the symlink's inode 1071 * @buf: a pointer to the buffer to be filled 1072 * @len: a pointer to the length of @buf 1073 * 1074 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1075 * to be freed by the caller. 1076 * 1077 * Returns: errno 1078 */ 1079 1080 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1081 { 1082 struct gfs2_holder i_gh; 1083 struct buffer_head *dibh; 1084 unsigned int x; 1085 int error; 1086 1087 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1088 error = gfs2_glock_nq_atime(&i_gh); 1089 if (error) { 1090 gfs2_holder_uninit(&i_gh); 1091 return error; 1092 } 1093 1094 if (!ip->i_di.di_size) { 1095 gfs2_consist_inode(ip); 1096 error = -EIO; 1097 goto out; 1098 } 1099 1100 error = gfs2_meta_inode_buffer(ip, &dibh); 1101 if (error) 1102 goto out; 1103 1104 x = ip->i_di.di_size + 1; 1105 if (x > *len) { 1106 *buf = kmalloc(x, GFP_KERNEL); 1107 if (!*buf) { 1108 error = -ENOMEM; 1109 goto out_brelse; 1110 } 1111 } 1112 1113 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1114 *len = x; 1115 1116 out_brelse: 1117 brelse(dibh); 1118 out: 1119 gfs2_glock_dq_uninit(&i_gh); 1120 return error; 1121 } 1122 1123 /** 1124 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1125 * conditionally update the inode's atime 1126 * @gh: the holder to acquire 1127 * 1128 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1129 * Update if the difference between the current time and the inode's current 1130 * atime is greater than an interval specified at mount. 1131 * 1132 * Returns: errno 1133 */ 1134 1135 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1136 { 1137 struct gfs2_glock *gl = gh->gh_gl; 1138 struct gfs2_sbd *sdp = gl->gl_sbd; 1139 struct gfs2_inode *ip = gl->gl_object; 1140 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1141 unsigned int state; 1142 int flags; 1143 int error; 1144 1145 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1146 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1147 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1148 return -EINVAL; 1149 1150 state = gh->gh_state; 1151 flags = gh->gh_flags; 1152 1153 error = gfs2_glock_nq(gh); 1154 if (error) 1155 return error; 1156 1157 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1158 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1159 return 0; 1160 1161 curtime = get_seconds(); 1162 if (curtime - ip->i_di.di_atime >= quantum) { 1163 gfs2_glock_dq(gh); 1164 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1165 gh); 1166 error = gfs2_glock_nq(gh); 1167 if (error) 1168 return error; 1169 1170 /* Verify that atime hasn't been updated while we were 1171 trying to get exclusive lock. */ 1172 1173 curtime = get_seconds(); 1174 if (curtime - ip->i_di.di_atime >= quantum) { 1175 struct buffer_head *dibh; 1176 1177 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1178 if (error == -EROFS) 1179 return 0; 1180 if (error) 1181 goto fail; 1182 1183 error = gfs2_meta_inode_buffer(ip, &dibh); 1184 if (error) 1185 goto fail_end_trans; 1186 1187 ip->i_di.di_atime = curtime; 1188 1189 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1190 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1191 brelse(dibh); 1192 1193 gfs2_trans_end(sdp); 1194 } 1195 1196 /* If someone else has asked for the glock, 1197 unlock and let them have it. Then reacquire 1198 in the original state. */ 1199 if (gfs2_glock_is_blocking(gl)) { 1200 gfs2_glock_dq(gh); 1201 gfs2_holder_reinit(state, flags, gh); 1202 return gfs2_glock_nq(gh); 1203 } 1204 } 1205 1206 return 0; 1207 1208 fail_end_trans: 1209 gfs2_trans_end(sdp); 1210 fail: 1211 gfs2_glock_dq(gh); 1212 return error; 1213 } 1214 1215 /** 1216 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1217 * @arg_a: the first structure 1218 * @arg_b: the second structure 1219 * 1220 * Returns: 1 if A > B 1221 * -1 if A < B 1222 * 0 if A = B 1223 */ 1224 1225 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1226 { 1227 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a; 1228 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b; 1229 struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1230 struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1231 int ret = 0; 1232 1233 if (a->ln_number > b->ln_number) 1234 ret = 1; 1235 else if (a->ln_number < b->ln_number) 1236 ret = -1; 1237 else { 1238 if (gh_a->gh_state == LM_ST_SHARED && 1239 gh_b->gh_state == LM_ST_EXCLUSIVE) 1240 ret = 1; 1241 else if (gh_a->gh_state == LM_ST_SHARED && 1242 (gh_b->gh_flags & GL_ATIME)) 1243 ret = 1; 1244 } 1245 1246 return ret; 1247 } 1248 1249 /** 1250 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1251 * atime update 1252 * @num_gh: the number of structures 1253 * @ghs: an array of struct gfs2_holder structures 1254 * 1255 * Returns: 0 on success (all glocks acquired), 1256 * errno on failure (no glocks acquired) 1257 */ 1258 1259 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1260 { 1261 struct gfs2_holder **p; 1262 unsigned int x; 1263 int error = 0; 1264 1265 if (!num_gh) 1266 return 0; 1267 1268 if (num_gh == 1) { 1269 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1270 if (ghs->gh_flags & GL_ATIME) 1271 error = gfs2_glock_nq_atime(ghs); 1272 else 1273 error = gfs2_glock_nq(ghs); 1274 return error; 1275 } 1276 1277 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1278 if (!p) 1279 return -ENOMEM; 1280 1281 for (x = 0; x < num_gh; x++) 1282 p[x] = &ghs[x]; 1283 1284 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1285 1286 for (x = 0; x < num_gh; x++) { 1287 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1288 1289 if (p[x]->gh_flags & GL_ATIME) 1290 error = gfs2_glock_nq_atime(p[x]); 1291 else 1292 error = gfs2_glock_nq(p[x]); 1293 1294 if (error) { 1295 while (x--) 1296 gfs2_glock_dq(p[x]); 1297 break; 1298 } 1299 } 1300 1301 kfree(p); 1302 1303 return error; 1304 } 1305 1306 1307 static int 1308 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1309 { 1310 struct buffer_head *dibh; 1311 int error; 1312 1313 error = gfs2_meta_inode_buffer(ip, &dibh); 1314 if (!error) { 1315 error = inode_setattr(&ip->i_inode, attr); 1316 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1317 gfs2_inode_attr_out(ip); 1318 1319 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1320 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1321 brelse(dibh); 1322 } 1323 return error; 1324 } 1325 1326 /** 1327 * gfs2_setattr_simple - 1328 * @ip: 1329 * @attr: 1330 * 1331 * Called with a reference on the vnode. 1332 * 1333 * Returns: errno 1334 */ 1335 1336 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1337 { 1338 int error; 1339 1340 if (current->journal_info) 1341 return __gfs2_setattr_simple(ip, attr); 1342 1343 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1344 if (error) 1345 return error; 1346 1347 error = __gfs2_setattr_simple(ip, attr); 1348 1349 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1350 1351 return error; 1352 } 1353 1354