fs/gfs2/inode.c

   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 static int iget_test(struct inode *inode, void *opaque)
  42 {
  43         struct gfs2_inode *ip = GFS2_I(inode);
  44         struct gfs2_inum_host *inum = opaque;
  45
  46         if (ip->i_num.no_addr == inum->no_addr)
  47                 return 1;
  48
  49         return 0;
  50 }
  51
  52 static int iget_set(struct inode *inode, void *opaque)
  53 {
  54         struct gfs2_inode *ip = GFS2_I(inode);
  55         struct gfs2_inum_host *inum = opaque;
  56
  57         ip->i_num = *inum;
  58         inode->i_ino = inum->no_addr;
  59         return 0;
  60 }
  61
  62 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum_host *inum)
  63 {
  64         return ilookup5(sb, (unsigned long)inum->no_formal_ino,
  65                         iget_test, inum);
  66 }
  67
  68 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum_host *inum)
  69 {
  70         return iget5_locked(sb, (unsigned long)inum->no_formal_ino,
  71                      iget_test, iget_set, inum);
  72 }
  73
  74 /**
  75  * gfs2_inode_lookup - Lookup an inode
  76  * @sb: The super block
  77  * @inum: The inode number
  78  * @type: The type of the inode
  79  *
  80  * Returns: A VFS inode, or an error
  81  */
  82
  83 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum_host *inum, unsigned int type)
  84 {
  85         struct inode *inode = gfs2_iget(sb, inum);
  86         struct gfs2_inode *ip = GFS2_I(inode);
  87         struct gfs2_glock *io_gl;
  88         int error;
  89
  90         if (!inode)
  91                 return ERR_PTR(-ENOBUFS);
  92
  93         if (inode->i_state & I_NEW) {
  94                 struct gfs2_sbd *sdp = GFS2_SB(inode);
  95                 umode_t mode = DT2IF(type);
  96                 inode->i_private = ip;
  97                 inode->i_mode = mode;
  98
  99                 if (S_ISREG(mode)) {
 100                         inode->i_op = &gfs2_file_iops;
 101                         inode->i_fop = &gfs2_file_fops;
 102                         inode->i_mapping->a_ops = &gfs2_file_aops;
 103                 } else if (S_ISDIR(mode)) {
 104                         inode->i_op = &gfs2_dir_iops;
 105                         inode->i_fop = &gfs2_dir_fops;
 106                 } else if (S_ISLNK(mode)) {
 107                         inode->i_op = &gfs2_symlink_iops;
 108                 } else {
 109                         inode->i_op = &gfs2_dev_iops;
 110                 }
 111
 112                 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
 113                 if (unlikely(error))
 114                         goto fail;
 115                 ip->i_gl->gl_object = ip;
 116
 117                 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
 118                 if (unlikely(error))
 119                         goto fail_put;
 120
 121                 set_bit(GIF_INVALID, &ip->i_flags);
 122                 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
 123                 if (unlikely(error))
 124                         goto fail_iopen;
 125
 126                 gfs2_glock_put(io_gl);
 127                 unlock_new_inode(inode);
 128         }
 129
 130         return inode;
 131 fail_iopen:
 132         gfs2_glock_put(io_gl);
 133 fail_put:
 134         ip->i_gl->gl_object = NULL;
 135         gfs2_glock_put(ip->i_gl);
 136 fail:
 137         iput(inode);
 138         return ERR_PTR(error);
 139 }
 140
 141 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
 142 {
 143         struct gfs2_dinode_host *di = &ip->i_di;
 144         const struct gfs2_dinode *str = buf;
 145
 146         if (ip->i_num.no_addr != be64_to_cpu(str->di_num.no_addr)) {
 147                 if (gfs2_consist_inode(ip))
 148                         gfs2_dinode_print(ip);
 149                 return -EIO;
 150         }
 151         if (ip->i_num.no_formal_ino != be64_to_cpu(str->di_num.no_formal_ino))
 152                 return -ESTALE;
 153
 154         ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
 155         ip->i_inode.i_rdev = 0;
 156         switch (ip->i_inode.i_mode & S_IFMT) {
 157         case S_IFBLK:
 158         case S_IFCHR:
 159                 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
 160                                            be32_to_cpu(str->di_minor));
 161                 break;
 162         };
 163
 164         ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
 165         ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
 166         /*
 167          * We will need to review setting the nlink count here in the
 168          * light of the forthcoming ro bind mount work. This is a reminder
 169          * to do that.
 170          */
 171         ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink);
 172         di->di_size = be64_to_cpu(str->di_size);
 173         i_size_write(&ip->i_inode, di->di_size);
 174         di->di_blocks = be64_to_cpu(str->di_blocks);
 175         gfs2_set_inode_blocks(&ip->i_inode);
 176         ip->i_inode.i_atime.tv_sec = be64_to_cpu(str->di_atime);
 177         ip->i_inode.i_atime.tv_nsec = 0;
 178         ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
 179         ip->i_inode.i_mtime.tv_nsec = 0;
 180         ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
 181         ip->i_inode.i_ctime.tv_nsec = 0;
 182
 183         di->di_goal_meta = be64_to_cpu(str->di_goal_meta);
 184         di->di_goal_data = be64_to_cpu(str->di_goal_data);
 185         di->di_generation = be64_to_cpu(str->di_generation);
 186
 187         di->di_flags = be32_to_cpu(str->di_flags);
 188         gfs2_set_inode_flags(&ip->i_inode);
 189         di->di_height = be16_to_cpu(str->di_height);
 190
 191         di->di_depth = be16_to_cpu(str->di_depth);
 192         di->di_entries = be32_to_cpu(str->di_entries);
 193
 194         di->di_eattr = be64_to_cpu(str->di_eattr);
 195         return 0;
 196 }
 197
 198 /**
 199  * gfs2_inode_refresh - Refresh the incore copy of the dinode
 200  * @ip: The GFS2 inode
 201  *
 202  * Returns: errno
 203  */
 204
 205 int gfs2_inode_refresh(struct gfs2_inode *ip)
 206 {
 207         struct buffer_head *dibh;
 208         int error;
 209
 210         error = gfs2_meta_inode_buffer(ip, &dibh);
 211         if (error)
 212                 return error;
 213
 214         if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
 215                 brelse(dibh);
 216                 return -EIO;
 217         }
 218
 219         error = gfs2_dinode_in(ip, dibh->b_data);
 220         brelse(dibh);
 221         clear_bit(GIF_INVALID, &ip->i_flags);
 222
 223         return error;
 224 }
 225
 226 int gfs2_dinode_dealloc(struct gfs2_inode *ip)
 227 {
 228         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 229         struct gfs2_alloc *al;
 230         struct gfs2_rgrpd *rgd;
 231         int error;
 232
 233         if (ip->i_di.di_blocks != 1) {
 234                 if (gfs2_consist_inode(ip))
 235                         gfs2_dinode_print(ip);
 236                 return -EIO;
 237         }
 238
 239         al = gfs2_alloc_get(ip);
 240
 241         error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
 242         if (error)
 243                 goto out;
 244
 245         error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
 246         if (error)
 247                 goto out_qs;
 248
 249         rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
 250         if (!rgd) {
 251                 gfs2_consist_inode(ip);
 252                 error = -EIO;
 253                 goto out_rindex_relse;
 254         }
 255
 256         error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
 257                                    &al->al_rgd_gh);
 258         if (error)
 259                 goto out_rindex_relse;
 260
 261         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1);
 262         if (error)
 263                 goto out_rg_gunlock;
 264
 265         gfs2_trans_add_gl(ip->i_gl);
 266
 267         gfs2_free_di(rgd, ip);
 268
 269         gfs2_trans_end(sdp);
 270         clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
 271
 272 out_rg_gunlock:
 273         gfs2_glock_dq_uninit(&al->al_rgd_gh);
 274 out_rindex_relse:
 275         gfs2_glock_dq_uninit(&al->al_ri_gh);
 276 out_qs:
 277         gfs2_quota_unhold(ip);
 278 out:
 279         gfs2_alloc_put(ip);
 280         return error;
 281 }
 282
 283 /**
 284  * gfs2_change_nlink - Change nlink count on inode
 285  * @ip: The GFS2 inode
 286  * @diff: The change in the nlink count required
 287  *
 288  * Returns: errno
 289  */
 290
 291 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
 292 {
 293         struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info;
 294         struct buffer_head *dibh;
 295         u32 nlink;
 296         int error;
 297
 298         BUG_ON(diff != 1 && diff != -1);
 299         nlink = ip->i_inode.i_nlink + diff;
 300
 301         /* If we are reducing the nlink count, but the new value ends up being
 302            bigger than the old one, we must have underflowed. */
 303         if (diff < 0 && nlink > ip->i_inode.i_nlink) {
 304                 if (gfs2_consist_inode(ip))
 305                         gfs2_dinode_print(ip);
 306                 return -EIO;
 307         }
 308
 309         error = gfs2_meta_inode_buffer(ip, &dibh);
 310         if (error)
 311                 return error;
 312
 313         if (diff > 0)
 314                 inc_nlink(&ip->i_inode);
 315         else
 316                 drop_nlink(&ip->i_inode);
 317
 318         ip->i_inode.i_ctime.tv_sec = get_seconds();
 319
 320         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 321         gfs2_dinode_out(ip, dibh->b_data);
 322         brelse(dibh);
 323         mark_inode_dirty(&ip->i_inode);
 324
 325         if (ip->i_inode.i_nlink == 0) {
 326                 struct gfs2_rgrpd *rgd;
 327                 struct gfs2_holder ri_gh, rg_gh;
 328
 329                 error = gfs2_rindex_hold(sdp, &ri_gh);
 330                 if (error)
 331                         goto out;
 332                 error = -EIO;
 333                 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
 334                 if (!rgd)
 335                         goto out_norgrp;
 336                 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh);
 337                 if (error)
 338                         goto out_norgrp;
 339
 340                 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */
 341                 gfs2_glock_dq_uninit(&rg_gh);
 342 out_norgrp:
 343                 gfs2_glock_dq_uninit(&ri_gh);
 344         }
 345 out:
 346         return error;
 347 }
 348
 349 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
 350 {
 351         struct qstr qstr;
 352         gfs2_str2qstr(&qstr, name);
 353         return gfs2_lookupi(dip, &qstr, 1, NULL);
 354 }
 355
 356
 357 /**
 358  * gfs2_lookupi - Look up a filename in a directory and return its inode
 359  * @d_gh: An initialized holder for the directory glock
 360  * @name: The name of the inode to look for
 361  * @is_root: If 1, ignore the caller's permissions
 362  * @i_gh: An uninitialized holder for the new inode glock
 363  *
 364  * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
 365  * @is_root is true.
 366  *
 367  * Returns: errno
 368  */
 369
 370 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
 371                            int is_root, struct nameidata *nd)
 372 {
 373         struct super_block *sb = dir->i_sb;
 374         struct gfs2_inode *dip = GFS2_I(dir);
 375         struct gfs2_holder d_gh;
 376         struct gfs2_inum_host inum;
 377         unsigned int type;
 378         int error = 0;
 379         struct inode *inode = NULL;
 380
 381         if (!name->len || name->len > GFS2_FNAMESIZE)
 382                 return ERR_PTR(-ENAMETOOLONG);
 383
 384         if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
 385             (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
 386              dir == sb->s_root->d_inode)) {
 387                 igrab(dir);
 388                 return dir;
 389         }
 390
 391         error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 392         if (error)
 393                 return ERR_PTR(error);
 394
 395         if (!is_root) {
 396                 error = permission(dir, MAY_EXEC, NULL);
 397                 if (error)
 398                         goto out;
 399         }
 400
 401         error = gfs2_dir_search(dir, name, &inum, &type);
 402         if (error)
 403                 goto out;
 404
 405         inode = gfs2_inode_lookup(sb, &inum, type);
 406
 407 out:
 408         gfs2_glock_dq_uninit(&d_gh);
 409         if (error == -ENOENT)
 410                 return NULL;
 411         return inode;
 412 }
 413
 414 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino)
 415 {
 416         struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
 417         struct buffer_head *bh;
 418         struct gfs2_inum_range_host ir;
 419         int error;
 420
 421         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 422         if (error)
 423                 return error;
 424         mutex_lock(&sdp->sd_inum_mutex);
 425
 426         error = gfs2_meta_inode_buffer(ip, &bh);
 427         if (error) {
 428                 mutex_unlock(&sdp->sd_inum_mutex);
 429                 gfs2_trans_end(sdp);
 430                 return error;
 431         }
 432
 433         gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
 434
 435         if (ir.ir_length) {
 436                 *formal_ino = ir.ir_start++;
 437                 ir.ir_length--;
 438                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 439                 gfs2_inum_range_out(&ir,
 440                                     bh->b_data + sizeof(struct gfs2_dinode));
 441                 brelse(bh);
 442                 mutex_unlock(&sdp->sd_inum_mutex);
 443                 gfs2_trans_end(sdp);
 444                 return 0;
 445         }
 446
 447         brelse(bh);
 448
 449         mutex_unlock(&sdp->sd_inum_mutex);
 450         gfs2_trans_end(sdp);
 451
 452         return 1;
 453 }
 454
 455 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino)
 456 {
 457         struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
 458         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode);
 459         struct gfs2_holder gh;
 460         struct buffer_head *bh;
 461         struct gfs2_inum_range_host ir;
 462         int error;
 463
 464         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 465         if (error)
 466                 return error;
 467
 468         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 469         if (error)
 470                 goto out;
 471         mutex_lock(&sdp->sd_inum_mutex);
 472
 473         error = gfs2_meta_inode_buffer(ip, &bh);
 474         if (error)
 475                 goto out_end_trans;
 476
 477         gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
 478
 479         if (!ir.ir_length) {
 480                 struct buffer_head *m_bh;
 481                 u64 x, y;
 482                 __be64 z;
 483
 484                 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 485                 if (error)
 486                         goto out_brelse;
 487
 488                 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode));
 489                 x = y = be64_to_cpu(z);
 490                 ir.ir_start = x;
 491                 ir.ir_length = GFS2_INUM_QUANTUM;
 492                 x += GFS2_INUM_QUANTUM;
 493                 if (x < y)
 494                         gfs2_consist_inode(m_ip);
 495                 z = cpu_to_be64(x);
 496                 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 497                 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z;
 498
 499                 brelse(m_bh);
 500         }
 501
 502         *formal_ino = ir.ir_start++;
 503         ir.ir_length--;
 504
 505         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 506         gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
 507
 508 out_brelse:
 509         brelse(bh);
 510 out_end_trans:
 511         mutex_unlock(&sdp->sd_inum_mutex);
 512         gfs2_trans_end(sdp);
 513 out:
 514         gfs2_glock_dq_uninit(&gh);
 515         return error;
 516 }
 517
 518 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum)
 519 {
 520         int error;
 521
 522         error = pick_formal_ino_1(sdp, inum);
 523         if (error <= 0)
 524                 return error;
 525
 526         error = pick_formal_ino_2(sdp, inum);
 527
 528         return error;
 529 }
 530
 531 /**
 532  * create_ok - OK to create a new on-disk inode here?
 533  * @dip:  Directory in which dinode is to be created
 534  * @name:  Name of new dinode
 535  * @mode:
 536  *
 537  * Returns: errno
 538  */
 539
 540 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
 541                      unsigned int mode)
 542 {
 543         int error;
 544
 545         error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
 546         if (error)
 547                 return error;
 548
 549         /*  Don't create entries in an unlinked directory  */
 550         if (!dip->i_inode.i_nlink)
 551                 return -EPERM;
 552
 553         error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL);
 554         switch (error) {
 555         case -ENOENT:
 556                 error = 0;
 557                 break;
 558         case 0:
 559                 return -EEXIST;
 560         default:
 561                 return error;
 562         }
 563
 564         if (dip->i_di.di_entries == (u32)-1)
 565                 return -EFBIG;
 566         if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
 567                 return -EMLINK;
 568
 569         return 0;
 570 }
 571
 572 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
 573                                unsigned int *uid, unsigned int *gid)
 574 {
 575         if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
 576             (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
 577                 if (S_ISDIR(*mode))
 578                         *mode |= S_ISUID;
 579                 else if (dip->i_inode.i_uid != current->fsuid)
 580                         *mode &= ~07111;
 581                 *uid = dip->i_inode.i_uid;
 582         } else
 583                 *uid = current->fsuid;
 584
 585         if (dip->i_inode.i_mode & S_ISGID) {
 586                 if (S_ISDIR(*mode))
 587                         *mode |= S_ISGID;
 588                 *gid = dip->i_inode.i_gid;
 589         } else
 590                 *gid = current->fsgid;
 591 }
 592
 593 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum,
 594                         u64 *generation)
 595 {
 596         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 597         int error;
 598
 599         gfs2_alloc_get(dip);
 600
 601         dip->i_alloc.al_requested = RES_DINODE;
 602         error = gfs2_inplace_reserve(dip);
 603         if (error)
 604                 goto out;
 605
 606         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
 607         if (error)
 608                 goto out_ipreserv;
 609
 610         inum->no_addr = gfs2_alloc_di(dip, generation);
 611
 612         gfs2_trans_end(sdp);
 613
 614 out_ipreserv:
 615         gfs2_inplace_release(dip);
 616 out:
 617         gfs2_alloc_put(dip);
 618         return error;
 619 }
 620
 621 /**
 622  * init_dinode - Fill in a new dinode structure
 623  * @dip: the directory this inode is being created in
 624  * @gl: The glock covering the new inode
 625  * @inum: the inode number
 626  * @mode: the file permissions
 627  * @uid:
 628  * @gid:
 629  *
 630  */
 631
 632 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
 633                         const struct gfs2_inum_host *inum, unsigned int mode,
 634                         unsigned int uid, unsigned int gid,
 635                         const u64 *generation, dev_t dev)
 636 {
 637         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 638         struct gfs2_dinode *di;
 639         struct buffer_head *dibh;
 640
 641         dibh = gfs2_meta_new(gl, inum->no_addr);
 642         gfs2_trans_add_bh(gl, dibh, 1);
 643         gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
 644         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 645         di = (struct gfs2_dinode *)dibh->b_data;
 646
 647         di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
 648         di->di_num.no_addr = cpu_to_be64(inum->no_addr);
 649         di->di_mode = cpu_to_be32(mode);
 650         di->di_uid = cpu_to_be32(uid);
 651         di->di_gid = cpu_to_be32(gid);
 652         di->di_nlink = 0;
 653         di->di_size = 0;
 654         di->di_blocks = cpu_to_be64(1);
 655         di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
 656         di->di_major = cpu_to_be32(MAJOR(dev));
 657         di->di_minor = cpu_to_be32(MINOR(dev));
 658         di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
 659         di->di_generation = cpu_to_be64(*generation);
 660         di->di_flags = 0;
 661
 662         if (S_ISREG(mode)) {
 663                 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
 664                     gfs2_tune_get(sdp, gt_new_files_jdata))
 665                         di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
 666                 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
 667                     gfs2_tune_get(sdp, gt_new_files_directio))
 668                         di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
 669         } else if (S_ISDIR(mode)) {
 670                 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
 671                                             GFS2_DIF_INHERIT_DIRECTIO);
 672                 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
 673                                             GFS2_DIF_INHERIT_JDATA);
 674         }
 675
 676         di->__pad1 = 0;
 677         di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
 678         di->di_height = 0;
 679         di->__pad2 = 0;
 680         di->__pad3 = 0;
 681         di->di_depth = 0;
 682         di->di_entries = 0;
 683         memset(&di->__pad4, 0, sizeof(di->__pad4));
 684         di->di_eattr = 0;
 685         memset(&di->di_reserved, 0, sizeof(di->di_reserved));
 686
 687         brelse(dibh);
 688 }
 689
 690 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
 691                        unsigned int mode, const struct gfs2_inum_host *inum,
 692                        const u64 *generation, dev_t dev)
 693 {
 694         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 695         unsigned int uid, gid;
 696         int error;
 697
 698         munge_mode_uid_gid(dip, &mode, &uid, &gid);
 699         gfs2_alloc_get(dip);
 700
 701         error = gfs2_quota_lock(dip, uid, gid);
 702         if (error)
 703                 goto out;
 704
 705         error = gfs2_quota_check(dip, uid, gid);
 706         if (error)
 707                 goto out_quota;
 708
 709         error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
 710         if (error)
 711                 goto out_quota;
 712
 713         init_dinode(dip, gl, inum, mode, uid, gid, generation, dev);
 714         gfs2_quota_change(dip, +1, uid, gid);
 715         gfs2_trans_end(sdp);
 716
 717 out_quota:
 718         gfs2_quota_unlock(dip);
 719 out:
 720         gfs2_alloc_put(dip);
 721         return error;
 722 }
 723
 724 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
 725                        struct gfs2_inode *ip)
 726 {
 727         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 728         struct gfs2_alloc *al;
 729         int alloc_required;
 730         struct buffer_head *dibh;
 731         int error;
 732
 733         al = gfs2_alloc_get(dip);
 734
 735         error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
 736         if (error)
 737                 goto fail;
 738
 739         error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
 740         if (alloc_required < 0)
 741                 goto fail;
 742         if (alloc_required) {
 743                 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
 744                 if (error)
 745                         goto fail_quota_locks;
 746
 747                 al->al_requested = sdp->sd_max_dirres;
 748
 749                 error = gfs2_inplace_reserve(dip);
 750                 if (error)
 751                         goto fail_quota_locks;
 752
 753                 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
 754                                          al->al_rgd->rd_ri.ri_length +
 755                                          2 * RES_DINODE +
 756                                          RES_STATFS + RES_QUOTA, 0);
 757                 if (error)
 758                         goto fail_ipreserv;
 759         } else {
 760                 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
 761                 if (error)
 762                         goto fail_quota_locks;
 763         }
 764
 765         error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_inode.i_mode));
 766         if (error)
 767                 goto fail_end_trans;
 768
 769         error = gfs2_meta_inode_buffer(ip, &dibh);
 770         if (error)
 771                 goto fail_end_trans;
 772         ip->i_inode.i_nlink = 1;
 773         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 774         gfs2_dinode_out(ip, dibh->b_data);
 775         brelse(dibh);
 776         return 0;
 777
 778 fail_end_trans:
 779         gfs2_trans_end(sdp);
 780
 781 fail_ipreserv:
 782         if (dip->i_alloc.al_rgd)
 783                 gfs2_inplace_release(dip);
 784
 785 fail_quota_locks:
 786         gfs2_quota_unlock(dip);
 787
 788 fail:
 789         gfs2_alloc_put(dip);
 790         return error;
 791 }
 792
 793 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip)
 794 {
 795         int err;
 796         size_t len;
 797         void *value;
 798         char *name;
 799         struct gfs2_ea_request er;
 800
 801         err = security_inode_init_security(&ip->i_inode, &dip->i_inode,
 802                                            &name, &value, &len);
 803
 804         if (err) {
 805                 if (err == -EOPNOTSUPP)
 806                         return 0;
 807                 return err;
 808         }
 809
 810         memset(&er, 0, sizeof(struct gfs2_ea_request));
 811
 812         er.er_type = GFS2_EATYPE_SECURITY;
 813         er.er_name = name;
 814         er.er_data = value;
 815         er.er_name_len = strlen(name);
 816         er.er_data_len = len;
 817
 818         err = gfs2_ea_set_i(ip, &er);
 819
 820         kfree(value);
 821         kfree(name);
 822
 823         return err;
 824 }
 825
 826 /**
 827  * gfs2_createi - Create a new inode
 828  * @ghs: An array of two holders
 829  * @name: The name of the new file
 830  * @mode: the permissions on the new inode
 831  *
 832  * @ghs[0] is an initialized holder for the directory
 833  * @ghs[1] is the holder for the inode lock
 834  *
 835  * If the return value is not NULL, the glocks on both the directory and the new
 836  * file are held.  A transaction has been started and an inplace reservation
 837  * is held, as well.
 838  *
 839  * Returns: An inode
 840  */
 841
 842 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name,
 843                            unsigned int mode, dev_t dev)
 844 {
 845         struct inode *inode;
 846         struct gfs2_inode *dip = ghs->gh_gl->gl_object;
 847         struct inode *dir = &dip->i_inode;
 848         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 849         struct gfs2_inum_host inum;
 850         int error;
 851         u64 generation;
 852
 853         if (!name->len || name->len > GFS2_FNAMESIZE)
 854                 return ERR_PTR(-ENAMETOOLONG);
 855
 856         gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
 857         error = gfs2_glock_nq(ghs);
 858         if (error)
 859                 goto fail;
 860
 861         error = create_ok(dip, name, mode);
 862         if (error)
 863                 goto fail_gunlock;
 864
 865         error = pick_formal_ino(sdp, &inum.no_formal_ino);
 866         if (error)
 867                 goto fail_gunlock;
 868
 869         error = alloc_dinode(dip, &inum, &generation);
 870         if (error)
 871                 goto fail_gunlock;
 872
 873         error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
 874                                   LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
 875         if (error)
 876                 goto fail_gunlock;
 877
 878         error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev);
 879         if (error)
 880                 goto fail_gunlock2;
 881
 882         inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode));
 883         if (IS_ERR(inode))
 884                 goto fail_gunlock2;
 885
 886         error = gfs2_inode_refresh(GFS2_I(inode));
 887         if (error)
 888                 goto fail_iput;
 889
 890         error = gfs2_acl_create(dip, GFS2_I(inode));
 891         if (error)
 892                 goto fail_iput;
 893
 894         error = gfs2_security_init(dip, GFS2_I(inode));
 895         if (error)
 896                 goto fail_iput;
 897
 898         error = link_dinode(dip, name, GFS2_I(inode));
 899         if (error)
 900                 goto fail_iput;
 901
 902         if (!inode)
 903                 return ERR_PTR(-ENOMEM);
 904         return inode;
 905
 906 fail_iput:
 907         iput(inode);
 908 fail_gunlock2:
 909         gfs2_glock_dq_uninit(ghs + 1);
 910 fail_gunlock:
 911         gfs2_glock_dq(ghs);
 912 fail:
 913         return ERR_PTR(error);
 914 }
 915
 916 /**
 917  * gfs2_rmdiri - Remove a directory
 918  * @dip: The parent directory of the directory to be removed
 919  * @name: The name of the directory to be removed
 920  * @ip: The GFS2 inode of the directory to be removed
 921  *
 922  * Assumes Glocks on dip and ip are held
 923  *
 924  * Returns: errno
 925  */
 926
 927 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name,
 928                 struct gfs2_inode *ip)
 929 {
 930         struct qstr dotname;
 931         int error;
 932
 933         if (ip->i_di.di_entries != 2) {
 934                 if (gfs2_consist_inode(ip))
 935                         gfs2_dinode_print(ip);
 936                 return -EIO;
 937         }
 938
 939         error = gfs2_dir_del(dip, name);
 940         if (error)
 941                 return error;
 942
 943         error = gfs2_change_nlink(dip, -1);
 944         if (error)
 945                 return error;
 946
 947         gfs2_str2qstr(&dotname, ".");
 948         error = gfs2_dir_del(ip, &dotname);
 949         if (error)
 950                 return error;
 951
 952         gfs2_str2qstr(&dotname, "..");
 953         error = gfs2_dir_del(ip, &dotname);
 954         if (error)
 955                 return error;
 956
 957         /* It looks odd, but it really should be done twice */
 958         error = gfs2_change_nlink(ip, -1);
 959         if (error)
 960                 return error;
 961
 962         error = gfs2_change_nlink(ip, -1);
 963         if (error)
 964                 return error;
 965
 966         return error;
 967 }
 968
 969 /*
 970  * gfs2_unlink_ok - check to see that a inode is still in a directory
 971  * @dip: the directory
 972  * @name: the name of the file
 973  * @ip: the inode
 974  *
 975  * Assumes that the lock on (at least) @dip is held.
 976  *
 977  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
 978  */
 979
 980 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
 981                    struct gfs2_inode *ip)
 982 {
 983         struct gfs2_inum_host inum;
 984         unsigned int type;
 985         int error;
 986
 987         if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
 988                 return -EPERM;
 989
 990         if ((dip->i_inode.i_mode & S_ISVTX) &&
 991             dip->i_inode.i_uid != current->fsuid &&
 992             ip->i_inode.i_uid != current->fsuid && !capable(CAP_FOWNER))
 993                 return -EPERM;
 994
 995         if (IS_APPEND(&dip->i_inode))
 996                 return -EPERM;
 997
 998         error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
 999         if (error)
1000                 return error;
1001
1002         error = gfs2_dir_search(&dip->i_inode, name, &inum, &type);
1003         if (error)
1004                 return error;
1005
1006         if (!gfs2_inum_equal(&inum, &ip->i_num))
1007                 return -ENOENT;
1008
1009         if (IF2DT(ip->i_inode.i_mode) != type) {
1010                 gfs2_consist_inode(dip);
1011                 return -EIO;
1012         }
1013
1014         return 0;
1015 }
1016
1017 /*
1018  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1019  * @this: move this
1020  * @to: to here
1021  *
1022  * Follow @to back to the root and make sure we don't encounter @this
1023  * Assumes we already hold the rename lock.
1024  *
1025  * Returns: errno
1026  */
1027
1028 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1029 {
1030         struct inode *dir = &to->i_inode;
1031         struct super_block *sb = dir->i_sb;
1032         struct inode *tmp;
1033         struct qstr dotdot;
1034         int error = 0;
1035
1036         gfs2_str2qstr(&dotdot, "..");
1037
1038         igrab(dir);
1039
1040         for (;;) {
1041                 if (dir == &this->i_inode) {
1042                         error = -EINVAL;
1043                         break;
1044                 }
1045                 if (dir == sb->s_root->d_inode) {
1046                         error = 0;
1047                         break;
1048                 }
1049
1050                 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1051                 if (IS_ERR(tmp)) {
1052                         error = PTR_ERR(tmp);
1053                         break;
1054                 }
1055
1056                 iput(dir);
1057                 dir = tmp;
1058         }
1059
1060         iput(dir);
1061
1062         return error;
1063 }
1064
1065 /**
1066  * gfs2_readlinki - return the contents of a symlink
1067  * @ip: the symlink's inode
1068  * @buf: a pointer to the buffer to be filled
1069  * @len: a pointer to the length of @buf
1070  *
1071  * If @buf is too small, a piece of memory is kmalloc()ed and needs
1072  * to be freed by the caller.
1073  *
1074  * Returns: errno
1075  */
1076
1077 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1078 {
1079         struct gfs2_holder i_gh;
1080         struct buffer_head *dibh;
1081         unsigned int x;
1082         int error;
1083
1084         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1085         error = gfs2_glock_nq_atime(&i_gh);
1086         if (error) {
1087                 gfs2_holder_uninit(&i_gh);
1088                 return error;
1089         }
1090
1091         if (!ip->i_di.di_size) {
1092                 gfs2_consist_inode(ip);
1093                 error = -EIO;
1094                 goto out;
1095         }
1096
1097         error = gfs2_meta_inode_buffer(ip, &dibh);
1098         if (error)
1099                 goto out;
1100
1101         x = ip->i_di.di_size + 1;
1102         if (x > *len) {
1103                 *buf = kmalloc(x, GFP_KERNEL);
1104                 if (!*buf) {
1105                         error = -ENOMEM;
1106                         goto out_brelse;
1107                 }
1108         }
1109
1110         memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1111         *len = x;
1112
1113 out_brelse:
1114         brelse(dibh);
1115 out:
1116         gfs2_glock_dq_uninit(&i_gh);
1117         return error;
1118 }
1119
1120 /**
1121  * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1122  *       conditionally update the inode's atime
1123  * @gh: the holder to acquire
1124  *
1125  * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1126  * Update if the difference between the current time and the inode's current
1127  * atime is greater than an interval specified at mount.
1128  *
1129  * Returns: errno
1130  */
1131
1132 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1133 {
1134         struct gfs2_glock *gl = gh->gh_gl;
1135         struct gfs2_sbd *sdp = gl->gl_sbd;
1136         struct gfs2_inode *ip = gl->gl_object;
1137         s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1138         unsigned int state;
1139         int flags;
1140         int error;
1141
1142         if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1143             gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1144             gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1145                 return -EINVAL;
1146
1147         state = gh->gh_state;
1148         flags = gh->gh_flags;
1149
1150         error = gfs2_glock_nq(gh);
1151         if (error)
1152                 return error;
1153
1154         if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1155             (sdp->sd_vfs->s_flags & MS_RDONLY))
1156                 return 0;
1157
1158         curtime = get_seconds();
1159         if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) {
1160                 gfs2_glock_dq(gh);
1161                 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY,
1162                                    gh);
1163                 error = gfs2_glock_nq(gh);
1164                 if (error)
1165                         return error;
1166
1167                 /* Verify that atime hasn't been updated while we were
1168                    trying to get exclusive lock. */
1169
1170                 curtime = get_seconds();
1171                 if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) {
1172                         struct buffer_head *dibh;
1173                         struct gfs2_dinode *di;
1174
1175                         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1176                         if (error == -EROFS)
1177                                 return 0;
1178                         if (error)
1179                                 goto fail;
1180
1181                         error = gfs2_meta_inode_buffer(ip, &dibh);
1182                         if (error)
1183                                 goto fail_end_trans;
1184
1185                         ip->i_inode.i_atime.tv_sec = curtime;
1186
1187                         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1188                         di = (struct gfs2_dinode *)dibh->b_data;
1189                         di->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
1190                         brelse(dibh);
1191
1192                         gfs2_trans_end(sdp);
1193                 }
1194
1195                 /* If someone else has asked for the glock,
1196                    unlock and let them have it. Then reacquire
1197                    in the original state. */
1198                 if (gfs2_glock_is_blocking(gl)) {
1199                         gfs2_glock_dq(gh);
1200                         gfs2_holder_reinit(state, flags, gh);
1201                         return gfs2_glock_nq(gh);
1202                 }
1203         }
1204
1205         return 0;
1206
1207 fail_end_trans:
1208         gfs2_trans_end(sdp);
1209 fail:
1210         gfs2_glock_dq(gh);
1211         return error;
1212 }
1213
1214 static int
1215 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1216 {
1217         struct buffer_head *dibh;
1218         int error;
1219
1220         error = gfs2_meta_inode_buffer(ip, &dibh);
1221         if (!error) {
1222                 error = inode_setattr(&ip->i_inode, attr);
1223                 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1224                 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1225                 gfs2_dinode_out(ip, dibh->b_data);
1226                 brelse(dibh);
1227         }
1228         return error;
1229 }
1230
1231 /**
1232  * gfs2_setattr_simple -
1233  * @ip:
1234  * @attr:
1235  *
1236  * Called with a reference on the vnode.
1237  *
1238  * Returns: errno
1239  */
1240
1241 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1242 {
1243         int error;
1244
1245         if (current->journal_info)
1246                 return __gfs2_setattr_simple(ip, attr);
1247
1248         error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
1249         if (error)
1250                 return error;
1251
1252         error = __gfs2_setattr_simple(ip, attr);
1253         gfs2_trans_end(GFS2_SB(&ip->i_inode));
1254         return error;
1255 }
1256