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 + RES_STATFS + RES_QUOTA, 1); 281 if (error) 282 goto out_rg_gunlock; 283 284 gfs2_trans_add_gl(ip->i_gl); 285 286 gfs2_free_di(rgd, ip); 287 288 gfs2_trans_end(sdp); 289 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 290 291 out_rg_gunlock: 292 gfs2_glock_dq_uninit(&al->al_rgd_gh); 293 out_rindex_relse: 294 gfs2_glock_dq_uninit(&al->al_ri_gh); 295 out_qs: 296 gfs2_quota_unhold(ip); 297 out: 298 gfs2_alloc_put(ip); 299 return error; 300 } 301 302 /** 303 * gfs2_change_nlink - Change nlink count on inode 304 * @ip: The GFS2 inode 305 * @diff: The change in the nlink count required 306 * 307 * Returns: errno 308 */ 309 310 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 311 { 312 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info; 313 struct buffer_head *dibh; 314 uint32_t nlink; 315 int error; 316 317 BUG_ON(ip->i_di.di_nlink != ip->i_inode.i_nlink); 318 nlink = ip->i_di.di_nlink + diff; 319 320 /* If we are reducing the nlink count, but the new value ends up being 321 bigger than the old one, we must have underflowed. */ 322 if (diff < 0 && nlink > ip->i_di.di_nlink) { 323 if (gfs2_consist_inode(ip)) 324 gfs2_dinode_print(&ip->i_di); 325 return -EIO; 326 } 327 328 error = gfs2_meta_inode_buffer(ip, &dibh); 329 if (error) 330 return error; 331 332 ip->i_di.di_nlink = nlink; 333 ip->i_di.di_ctime = get_seconds(); 334 ip->i_inode.i_nlink = nlink; 335 336 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 337 gfs2_dinode_out(&ip->i_di, dibh->b_data); 338 brelse(dibh); 339 mark_inode_dirty(&ip->i_inode); 340 341 if (ip->i_di.di_nlink == 0) { 342 struct gfs2_rgrpd *rgd; 343 struct gfs2_holder ri_gh, rg_gh; 344 345 error = gfs2_rindex_hold(sdp, &ri_gh); 346 if (error) 347 goto out; 348 error = -EIO; 349 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 350 if (!rgd) 351 goto out_norgrp; 352 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh); 353 if (error) 354 goto out_norgrp; 355 356 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 357 gfs2_glock_dq_uninit(&rg_gh); 358 out_norgrp: 359 gfs2_glock_dq_uninit(&ri_gh); 360 } 361 out: 362 return error; 363 } 364 365 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 366 { 367 struct qstr qstr; 368 gfs2_str2qstr(&qstr, name); 369 return gfs2_lookupi(dip, &qstr, 1, NULL); 370 } 371 372 373 /** 374 * gfs2_lookupi - Look up a filename in a directory and return its inode 375 * @d_gh: An initialized holder for the directory glock 376 * @name: The name of the inode to look for 377 * @is_root: If 1, ignore the caller's permissions 378 * @i_gh: An uninitialized holder for the new inode glock 379 * 380 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 381 * @is_root is true. 382 * 383 * Returns: errno 384 */ 385 386 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 387 int is_root, struct nameidata *nd) 388 389 { 390 struct super_block *sb = dir->i_sb; 391 struct gfs2_inode *dip = GFS2_I(dir); 392 struct gfs2_holder d_gh; 393 struct gfs2_inum inum; 394 unsigned int type; 395 int error = 0; 396 struct inode *inode = NULL; 397 398 if (!name->len || name->len > GFS2_FNAMESIZE) 399 return ERR_PTR(-ENAMETOOLONG); 400 401 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 402 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 403 dir == sb->s_root->d_inode)) { 404 igrab(dir); 405 return dir; 406 } 407 408 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 409 if (error) 410 return ERR_PTR(error); 411 412 if (!is_root) { 413 error = permission(dir, MAY_EXEC, NULL); 414 if (error) 415 goto out; 416 } 417 418 error = gfs2_dir_search(dir, name, &inum, &type); 419 if (error) 420 goto out; 421 422 inode = gfs2_inode_lookup(sb, &inum, type); 423 424 out: 425 gfs2_glock_dq_uninit(&d_gh); 426 if (error == -ENOENT) 427 return NULL; 428 return inode; 429 } 430 431 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino) 432 { 433 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 434 struct buffer_head *bh; 435 struct gfs2_inum_range ir; 436 int error; 437 438 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 439 if (error) 440 return error; 441 mutex_lock(&sdp->sd_inum_mutex); 442 443 error = gfs2_meta_inode_buffer(ip, &bh); 444 if (error) { 445 mutex_unlock(&sdp->sd_inum_mutex); 446 gfs2_trans_end(sdp); 447 return error; 448 } 449 450 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 451 452 if (ir.ir_length) { 453 *formal_ino = ir.ir_start++; 454 ir.ir_length--; 455 gfs2_trans_add_bh(ip->i_gl, bh, 1); 456 gfs2_inum_range_out(&ir, 457 bh->b_data + sizeof(struct gfs2_dinode)); 458 brelse(bh); 459 mutex_unlock(&sdp->sd_inum_mutex); 460 gfs2_trans_end(sdp); 461 return 0; 462 } 463 464 brelse(bh); 465 466 mutex_unlock(&sdp->sd_inum_mutex); 467 gfs2_trans_end(sdp); 468 469 return 1; 470 } 471 472 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino) 473 { 474 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 475 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 476 struct gfs2_holder gh; 477 struct buffer_head *bh; 478 struct gfs2_inum_range ir; 479 int error; 480 481 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 482 if (error) 483 return error; 484 485 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 486 if (error) 487 goto out; 488 mutex_lock(&sdp->sd_inum_mutex); 489 490 error = gfs2_meta_inode_buffer(ip, &bh); 491 if (error) 492 goto out_end_trans; 493 494 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 495 496 if (!ir.ir_length) { 497 struct buffer_head *m_bh; 498 uint64_t x, y; 499 500 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 501 if (error) 502 goto out_brelse; 503 504 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 505 x = y = be64_to_cpu(x); 506 ir.ir_start = x; 507 ir.ir_length = GFS2_INUM_QUANTUM; 508 x += GFS2_INUM_QUANTUM; 509 if (x < y) 510 gfs2_consist_inode(m_ip); 511 x = cpu_to_be64(x); 512 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 513 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x; 514 515 brelse(m_bh); 516 } 517 518 *formal_ino = ir.ir_start++; 519 ir.ir_length--; 520 521 gfs2_trans_add_bh(ip->i_gl, bh, 1); 522 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 523 524 out_brelse: 525 brelse(bh); 526 out_end_trans: 527 mutex_unlock(&sdp->sd_inum_mutex); 528 gfs2_trans_end(sdp); 529 out: 530 gfs2_glock_dq_uninit(&gh); 531 return error; 532 } 533 534 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum) 535 { 536 int error; 537 538 error = pick_formal_ino_1(sdp, inum); 539 if (error <= 0) 540 return error; 541 542 error = pick_formal_ino_2(sdp, inum); 543 544 return error; 545 } 546 547 /** 548 * create_ok - OK to create a new on-disk inode here? 549 * @dip: Directory in which dinode is to be created 550 * @name: Name of new dinode 551 * @mode: 552 * 553 * Returns: errno 554 */ 555 556 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 557 unsigned int mode) 558 { 559 int error; 560 561 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 562 if (error) 563 return error; 564 565 /* Don't create entries in an unlinked directory */ 566 if (!dip->i_di.di_nlink) 567 return -EPERM; 568 569 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL); 570 switch (error) { 571 case -ENOENT: 572 error = 0; 573 break; 574 case 0: 575 return -EEXIST; 576 default: 577 return error; 578 } 579 580 if (dip->i_di.di_entries == (uint32_t)-1) 581 return -EFBIG; 582 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1) 583 return -EMLINK; 584 585 return 0; 586 } 587 588 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 589 unsigned int *uid, unsigned int *gid) 590 { 591 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 592 (dip->i_di.di_mode & S_ISUID) && dip->i_di.di_uid) { 593 if (S_ISDIR(*mode)) 594 *mode |= S_ISUID; 595 else if (dip->i_di.di_uid != current->fsuid) 596 *mode &= ~07111; 597 *uid = dip->i_di.di_uid; 598 } else 599 *uid = current->fsuid; 600 601 if (dip->i_di.di_mode & S_ISGID) { 602 if (S_ISDIR(*mode)) 603 *mode |= S_ISGID; 604 *gid = dip->i_di.di_gid; 605 } else 606 *gid = current->fsgid; 607 } 608 609 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum *inum, 610 u64 *generation) 611 { 612 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 613 int error; 614 615 gfs2_alloc_get(dip); 616 617 dip->i_alloc.al_requested = RES_DINODE; 618 error = gfs2_inplace_reserve(dip); 619 if (error) 620 goto out; 621 622 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 623 if (error) 624 goto out_ipreserv; 625 626 inum->no_addr = gfs2_alloc_di(dip, generation); 627 628 gfs2_trans_end(sdp); 629 630 out_ipreserv: 631 gfs2_inplace_release(dip); 632 out: 633 gfs2_alloc_put(dip); 634 return error; 635 } 636 637 /** 638 * init_dinode - Fill in a new dinode structure 639 * @dip: the directory this inode is being created in 640 * @gl: The glock covering the new inode 641 * @inum: the inode number 642 * @mode: the file permissions 643 * @uid: 644 * @gid: 645 * 646 */ 647 648 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 649 const struct gfs2_inum *inum, unsigned int mode, 650 unsigned int uid, unsigned int gid, 651 const u64 *generation) 652 { 653 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 654 struct gfs2_dinode *di; 655 struct buffer_head *dibh; 656 657 dibh = gfs2_meta_new(gl, inum->no_addr); 658 gfs2_trans_add_bh(gl, dibh, 1); 659 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 660 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 661 di = (struct gfs2_dinode *)dibh->b_data; 662 663 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 664 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 665 di->di_mode = cpu_to_be32(mode); 666 di->di_uid = cpu_to_be32(uid); 667 di->di_gid = cpu_to_be32(gid); 668 di->di_nlink = cpu_to_be32(0); 669 di->di_size = cpu_to_be64(0); 670 di->di_blocks = cpu_to_be64(1); 671 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 672 di->di_major = di->di_minor = cpu_to_be32(0); 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 = cpu_to_be32(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(0); 693 di->di_height = cpu_to_be32(0); 694 di->__pad2 = 0; 695 di->__pad3 = 0; 696 di->di_depth = cpu_to_be16(0); 697 di->di_entries = cpu_to_be32(0); 698 memset(&di->__pad4, 0, sizeof(di->__pad4)); 699 di->di_eattr = cpu_to_be64(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 *inum, 707 const u64 *generation) 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); 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_di.di_uid, 759 dip->i_di.di_gid); 760 if (error) 761 goto fail_quota_locks; 762 763 al->al_requested = sdp->sd_max_dirres; 764 765 error = gfs2_inplace_reserve(dip); 766 if (error) 767 goto fail_quota_locks; 768 769 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 770 al->al_rgd->rd_ri.ri_length + 771 2 * RES_DINODE + 772 RES_STATFS + RES_QUOTA, 0); 773 if (error) 774 goto fail_ipreserv; 775 } else { 776 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 777 if (error) 778 goto fail_quota_locks; 779 } 780 781 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_di.di_mode)); 782 if (error) 783 goto fail_end_trans; 784 785 error = gfs2_meta_inode_buffer(ip, &dibh); 786 if (error) 787 goto fail_end_trans; 788 ip->i_di.di_nlink = 1; 789 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 790 gfs2_dinode_out(&ip->i_di, dibh->b_data); 791 brelse(dibh); 792 return 0; 793 794 fail_end_trans: 795 gfs2_trans_end(sdp); 796 797 fail_ipreserv: 798 if (dip->i_alloc.al_rgd) 799 gfs2_inplace_release(dip); 800 801 fail_quota_locks: 802 gfs2_quota_unlock(dip); 803 804 fail: 805 gfs2_alloc_put(dip); 806 return error; 807 } 808 809 /** 810 * gfs2_createi - Create a new inode 811 * @ghs: An array of two holders 812 * @name: The name of the new file 813 * @mode: the permissions on the new inode 814 * 815 * @ghs[0] is an initialized holder for the directory 816 * @ghs[1] is the holder for the inode lock 817 * 818 * If the return value is not NULL, the glocks on both the directory and the new 819 * file are held. A transaction has been started and an inplace reservation 820 * is held, as well. 821 * 822 * Returns: An inode 823 */ 824 825 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 826 unsigned int mode) 827 { 828 struct inode *inode; 829 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 830 struct inode *dir = &dip->i_inode; 831 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 832 struct gfs2_inum inum; 833 int error; 834 u64 generation; 835 836 if (!name->len || name->len > GFS2_FNAMESIZE) 837 return ERR_PTR(-ENAMETOOLONG); 838 839 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 840 error = gfs2_glock_nq(ghs); 841 if (error) 842 goto fail; 843 844 error = create_ok(dip, name, mode); 845 if (error) 846 goto fail_gunlock; 847 848 error = pick_formal_ino(sdp, &inum.no_formal_ino); 849 if (error) 850 goto fail_gunlock; 851 852 error = alloc_dinode(dip, &inum, &generation); 853 if (error) 854 goto fail_gunlock; 855 856 if (inum.no_addr < dip->i_num.no_addr) { 857 gfs2_glock_dq(ghs); 858 859 error = gfs2_glock_nq_num(sdp, inum.no_addr, 860 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 861 GL_SKIP, ghs + 1); 862 if (error) { 863 return ERR_PTR(error); 864 } 865 866 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 867 error = gfs2_glock_nq(ghs); 868 if (error) { 869 gfs2_glock_dq_uninit(ghs + 1); 870 return ERR_PTR(error); 871 } 872 873 error = create_ok(dip, name, mode); 874 if (error) 875 goto fail_gunlock2; 876 } else { 877 error = gfs2_glock_nq_num(sdp, inum.no_addr, 878 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 879 GL_SKIP, ghs + 1); 880 if (error) 881 goto fail_gunlock; 882 } 883 884 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation); 885 if (error) 886 goto fail_gunlock2; 887 888 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode)); 889 if (IS_ERR(inode)) 890 goto fail_gunlock2; 891 892 error = gfs2_inode_refresh(GFS2_I(inode)); 893 if (error) 894 goto fail_iput; 895 896 error = gfs2_acl_create(dip, GFS2_I(inode)); 897 if (error) 898 goto fail_iput; 899 900 error = link_dinode(dip, name, GFS2_I(inode)); 901 if (error) 902 goto fail_iput; 903 904 if (!inode) 905 return ERR_PTR(-ENOMEM); 906 return inode; 907 908 fail_iput: 909 iput(inode); 910 fail_gunlock2: 911 gfs2_glock_dq_uninit(ghs + 1); 912 fail_gunlock: 913 gfs2_glock_dq(ghs); 914 fail: 915 return ERR_PTR(error); 916 } 917 918 /** 919 * gfs2_rmdiri - Remove a directory 920 * @dip: The parent directory of the directory to be removed 921 * @name: The name of the directory to be removed 922 * @ip: The GFS2 inode of the directory to be removed 923 * 924 * Assumes Glocks on dip and ip are held 925 * 926 * Returns: errno 927 */ 928 929 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 930 struct gfs2_inode *ip) 931 { 932 struct qstr dotname; 933 int error; 934 935 if (ip->i_di.di_entries != 2) { 936 if (gfs2_consist_inode(ip)) 937 gfs2_dinode_print(&ip->i_di); 938 return -EIO; 939 } 940 941 error = gfs2_dir_del(dip, name); 942 if (error) 943 return error; 944 945 error = gfs2_change_nlink(dip, -1); 946 if (error) 947 return error; 948 949 gfs2_str2qstr(&dotname, "."); 950 error = gfs2_dir_del(ip, &dotname); 951 if (error) 952 return error; 953 954 gfs2_str2qstr(&dotname, ".."); 955 error = gfs2_dir_del(ip, &dotname); 956 if (error) 957 return error; 958 959 error = gfs2_change_nlink(ip, -2); 960 if (error) 961 return error; 962 963 return error; 964 } 965 966 /* 967 * gfs2_unlink_ok - check to see that a inode is still in a directory 968 * @dip: the directory 969 * @name: the name of the file 970 * @ip: the inode 971 * 972 * Assumes that the lock on (at least) @dip is held. 973 * 974 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 975 */ 976 977 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 978 struct gfs2_inode *ip) 979 { 980 struct gfs2_inum inum; 981 unsigned int type; 982 int error; 983 984 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 985 return -EPERM; 986 987 if ((dip->i_di.di_mode & S_ISVTX) && 988 dip->i_di.di_uid != current->fsuid && 989 ip->i_di.di_uid != current->fsuid && !capable(CAP_FOWNER)) 990 return -EPERM; 991 992 if (IS_APPEND(&dip->i_inode)) 993 return -EPERM; 994 995 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 996 if (error) 997 return error; 998 999 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type); 1000 if (error) 1001 return error; 1002 1003 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1004 return -ENOENT; 1005 1006 if (IF2DT(ip->i_di.di_mode) != type) { 1007 gfs2_consist_inode(dip); 1008 return -EIO; 1009 } 1010 1011 return 0; 1012 } 1013 1014 /* 1015 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1016 * @this: move this 1017 * @to: to here 1018 * 1019 * Follow @to back to the root and make sure we don't encounter @this 1020 * Assumes we already hold the rename lock. 1021 * 1022 * Returns: errno 1023 */ 1024 1025 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1026 { 1027 struct inode *dir = &to->i_inode; 1028 struct super_block *sb = dir->i_sb; 1029 struct inode *tmp; 1030 struct qstr dotdot; 1031 int error = 0; 1032 1033 gfs2_str2qstr(&dotdot, ".."); 1034 1035 igrab(dir); 1036 1037 for (;;) { 1038 if (dir == &this->i_inode) { 1039 error = -EINVAL; 1040 break; 1041 } 1042 if (dir == sb->s_root->d_inode) { 1043 error = 0; 1044 break; 1045 } 1046 1047 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1048 if (IS_ERR(tmp)) { 1049 error = PTR_ERR(tmp); 1050 break; 1051 } 1052 1053 iput(dir); 1054 dir = tmp; 1055 } 1056 1057 iput(dir); 1058 1059 return error; 1060 } 1061 1062 /** 1063 * gfs2_readlinki - return the contents of a symlink 1064 * @ip: the symlink's inode 1065 * @buf: a pointer to the buffer to be filled 1066 * @len: a pointer to the length of @buf 1067 * 1068 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1069 * to be freed by the caller. 1070 * 1071 * Returns: errno 1072 */ 1073 1074 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1075 { 1076 struct gfs2_holder i_gh; 1077 struct buffer_head *dibh; 1078 unsigned int x; 1079 int error; 1080 1081 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1082 error = gfs2_glock_nq_atime(&i_gh); 1083 if (error) { 1084 gfs2_holder_uninit(&i_gh); 1085 return error; 1086 } 1087 1088 if (!ip->i_di.di_size) { 1089 gfs2_consist_inode(ip); 1090 error = -EIO; 1091 goto out; 1092 } 1093 1094 error = gfs2_meta_inode_buffer(ip, &dibh); 1095 if (error) 1096 goto out; 1097 1098 x = ip->i_di.di_size + 1; 1099 if (x > *len) { 1100 *buf = kmalloc(x, GFP_KERNEL); 1101 if (!*buf) { 1102 error = -ENOMEM; 1103 goto out_brelse; 1104 } 1105 } 1106 1107 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1108 *len = x; 1109 1110 out_brelse: 1111 brelse(dibh); 1112 out: 1113 gfs2_glock_dq_uninit(&i_gh); 1114 return error; 1115 } 1116 1117 /** 1118 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1119 * conditionally update the inode's atime 1120 * @gh: the holder to acquire 1121 * 1122 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1123 * Update if the difference between the current time and the inode's current 1124 * atime is greater than an interval specified at mount. 1125 * 1126 * Returns: errno 1127 */ 1128 1129 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1130 { 1131 struct gfs2_glock *gl = gh->gh_gl; 1132 struct gfs2_sbd *sdp = gl->gl_sbd; 1133 struct gfs2_inode *ip = gl->gl_object; 1134 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1135 unsigned int state; 1136 int flags; 1137 int error; 1138 1139 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1140 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1141 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1142 return -EINVAL; 1143 1144 state = gh->gh_state; 1145 flags = gh->gh_flags; 1146 1147 error = gfs2_glock_nq(gh); 1148 if (error) 1149 return error; 1150 1151 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1152 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1153 return 0; 1154 1155 curtime = get_seconds(); 1156 if (curtime - ip->i_di.di_atime >= quantum) { 1157 gfs2_glock_dq(gh); 1158 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1159 gh); 1160 error = gfs2_glock_nq(gh); 1161 if (error) 1162 return error; 1163 1164 /* Verify that atime hasn't been updated while we were 1165 trying to get exclusive lock. */ 1166 1167 curtime = get_seconds(); 1168 if (curtime - ip->i_di.di_atime >= quantum) { 1169 struct buffer_head *dibh; 1170 1171 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1172 if (error == -EROFS) 1173 return 0; 1174 if (error) 1175 goto fail; 1176 1177 error = gfs2_meta_inode_buffer(ip, &dibh); 1178 if (error) 1179 goto fail_end_trans; 1180 1181 ip->i_di.di_atime = curtime; 1182 1183 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1184 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1185 brelse(dibh); 1186 1187 gfs2_trans_end(sdp); 1188 } 1189 1190 /* If someone else has asked for the glock, 1191 unlock and let them have it. Then reacquire 1192 in the original state. */ 1193 if (gfs2_glock_is_blocking(gl)) { 1194 gfs2_glock_dq(gh); 1195 gfs2_holder_reinit(state, flags, gh); 1196 return gfs2_glock_nq(gh); 1197 } 1198 } 1199 1200 return 0; 1201 1202 fail_end_trans: 1203 gfs2_trans_end(sdp); 1204 fail: 1205 gfs2_glock_dq(gh); 1206 return error; 1207 } 1208 1209 /** 1210 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1211 * @arg_a: the first structure 1212 * @arg_b: the second structure 1213 * 1214 * Returns: 1 if A > B 1215 * -1 if A < B 1216 * 0 if A = B 1217 */ 1218 1219 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1220 { 1221 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a; 1222 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b; 1223 struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1224 struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1225 int ret = 0; 1226 1227 if (a->ln_number > b->ln_number) 1228 ret = 1; 1229 else if (a->ln_number < b->ln_number) 1230 ret = -1; 1231 else { 1232 if (gh_a->gh_state == LM_ST_SHARED && 1233 gh_b->gh_state == LM_ST_EXCLUSIVE) 1234 ret = 1; 1235 else if (gh_a->gh_state == LM_ST_SHARED && 1236 (gh_b->gh_flags & GL_ATIME)) 1237 ret = 1; 1238 } 1239 1240 return ret; 1241 } 1242 1243 /** 1244 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1245 * atime update 1246 * @num_gh: the number of structures 1247 * @ghs: an array of struct gfs2_holder structures 1248 * 1249 * Returns: 0 on success (all glocks acquired), 1250 * errno on failure (no glocks acquired) 1251 */ 1252 1253 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1254 { 1255 struct gfs2_holder **p; 1256 unsigned int x; 1257 int error = 0; 1258 1259 if (!num_gh) 1260 return 0; 1261 1262 if (num_gh == 1) { 1263 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1264 if (ghs->gh_flags & GL_ATIME) 1265 error = gfs2_glock_nq_atime(ghs); 1266 else 1267 error = gfs2_glock_nq(ghs); 1268 return error; 1269 } 1270 1271 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1272 if (!p) 1273 return -ENOMEM; 1274 1275 for (x = 0; x < num_gh; x++) 1276 p[x] = &ghs[x]; 1277 1278 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1279 1280 for (x = 0; x < num_gh; x++) { 1281 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1282 1283 if (p[x]->gh_flags & GL_ATIME) 1284 error = gfs2_glock_nq_atime(p[x]); 1285 else 1286 error = gfs2_glock_nq(p[x]); 1287 1288 if (error) { 1289 while (x--) 1290 gfs2_glock_dq(p[x]); 1291 break; 1292 } 1293 } 1294 1295 kfree(p); 1296 return error; 1297 } 1298 1299 1300 static int 1301 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1302 { 1303 struct buffer_head *dibh; 1304 int error; 1305 1306 error = gfs2_meta_inode_buffer(ip, &dibh); 1307 if (!error) { 1308 error = inode_setattr(&ip->i_inode, attr); 1309 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1310 gfs2_inode_attr_out(ip); 1311 1312 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1313 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1314 brelse(dibh); 1315 } 1316 return error; 1317 } 1318 1319 /** 1320 * gfs2_setattr_simple - 1321 * @ip: 1322 * @attr: 1323 * 1324 * Called with a reference on the vnode. 1325 * 1326 * Returns: errno 1327 */ 1328 1329 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1330 { 1331 int error; 1332 1333 if (current->journal_info) 1334 return __gfs2_setattr_simple(ip, attr); 1335 1336 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1337 if (error) 1338 return error; 1339 1340 error = __gfs2_setattr_simple(ip, attr); 1341 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1342 return error; 1343 } 1344 1345