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 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
 291 {
 292         struct buffer_head *dibh;
 293         u32 nlink;
 294         int error;
 295
 296         BUG_ON(diff != 1 && diff != -1);
 297         nlink = ip->i_inode.i_nlink + diff;
 298
 299         /* If we are reducing the nlink count, but the new value ends up being
 300            bigger than the old one, we must have underflowed. */
 301         if (diff < 0 && nlink > ip->i_inode.i_nlink) {
 302                 if (gfs2_consist_inode(ip))
 303                         gfs2_dinode_print(ip);
 304                 return -EIO;
 305         }
 306
 307         error = gfs2_meta_inode_buffer(ip, &dibh);
 308         if (error)
 309                 return error;
 310
 311         if (diff > 0)
 312                 inc_nlink(&ip->i_inode);
 313         else
 314                 drop_nlink(&ip->i_inode);
 315
 316         ip->i_inode.i_ctime = CURRENT_TIME_SEC;
 317
 318         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 319         gfs2_dinode_out(ip, dibh->b_data);
 320         brelse(dibh);
 321         mark_inode_dirty(&ip->i_inode);
 322
 323         if (ip->i_inode.i_nlink == 0)
 324                 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */
 325
 326         return error;
 327 }
 328
 329 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
 330 {
 331         struct qstr qstr;
 332         struct inode *inode;
 333         gfs2_str2qstr(&qstr, name);
 334         inode = gfs2_lookupi(dip, &qstr, 1, NULL);
 335         /* gfs2_lookupi has inconsistent callers: vfs
 336          * related routines expect NULL for no entry found,
 337          * gfs2_lookup_simple callers expect ENOENT
 338          * and do not check for NULL.
 339          */
 340         if (inode == NULL)
 341                 return ERR_PTR(-ENOENT);
 342         else
 343                 return inode;
 344 }
 345
 346
 347 /**
 348  * gfs2_lookupi - Look up a filename in a directory and return its inode
 349  * @d_gh: An initialized holder for the directory glock
 350  * @name: The name of the inode to look for
 351  * @is_root: If 1, ignore the caller's permissions
 352  * @i_gh: An uninitialized holder for the new inode glock
 353  *
 354  * This can be called via the VFS filldir function when NFS is doing
 355  * a readdirplus and the inode which its intending to stat isn't
 356  * already in cache. In this case we must not take the directory glock
 357  * again, since the readdir call will have already taken that lock.
 358  *
 359  * Returns: errno
 360  */
 361
 362 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
 363                            int is_root, struct nameidata *nd)
 364 {
 365         struct super_block *sb = dir->i_sb;
 366         struct gfs2_inode *dip = GFS2_I(dir);
 367         struct gfs2_holder d_gh;
 368         struct gfs2_inum_host inum;
 369         unsigned int type;
 370         int error;
 371         struct inode *inode = NULL;
 372         int unlock = 0;
 373
 374         if (!name->len || name->len > GFS2_FNAMESIZE)
 375                 return ERR_PTR(-ENAMETOOLONG);
 376
 377         if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
 378             (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
 379              dir == sb->s_root->d_inode)) {
 380                 igrab(dir);
 381                 return dir;
 382         }
 383
 384         if (gfs2_glock_is_locked_by_me(dip->i_gl) == 0) {
 385                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 386                 if (error)
 387                         return ERR_PTR(error);
 388                 unlock = 1;
 389         }
 390
 391         if (!is_root) {
 392                 error = permission(dir, MAY_EXEC, NULL);
 393                 if (error)
 394                         goto out;
 395         }
 396
 397         error = gfs2_dir_search(dir, name, &inum, &type);
 398         if (error)
 399                 goto out;
 400
 401         inode = gfs2_inode_lookup(sb, &inum, type);
 402
 403 out:
 404         if (unlock)
 405                 gfs2_glock_dq_uninit(&d_gh);
 406         if (error == -ENOENT)
 407                 return NULL;
 408         return inode ? inode : ERR_PTR(error);
 409 }
 410
 411 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino)
 412 {
 413         struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
 414         struct buffer_head *bh;
 415         struct gfs2_inum_range_host ir;
 416         int error;
 417
 418         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 419         if (error)
 420                 return error;
 421         mutex_lock(&sdp->sd_inum_mutex);
 422
 423         error = gfs2_meta_inode_buffer(ip, &bh);
 424         if (error) {
 425                 mutex_unlock(&sdp->sd_inum_mutex);
 426                 gfs2_trans_end(sdp);
 427                 return error;
 428         }
 429
 430         gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
 431
 432         if (ir.ir_length) {
 433                 *formal_ino = ir.ir_start++;
 434                 ir.ir_length--;
 435                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 436                 gfs2_inum_range_out(&ir,
 437                                     bh->b_data + sizeof(struct gfs2_dinode));
 438                 brelse(bh);
 439                 mutex_unlock(&sdp->sd_inum_mutex);
 440                 gfs2_trans_end(sdp);
 441                 return 0;
 442         }
 443
 444         brelse(bh);
 445
 446         mutex_unlock(&sdp->sd_inum_mutex);
 447         gfs2_trans_end(sdp);
 448
 449         return 1;
 450 }
 451
 452 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino)
 453 {
 454         struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
 455         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode);
 456         struct gfs2_holder gh;
 457         struct buffer_head *bh;
 458         struct gfs2_inum_range_host ir;
 459         int error;
 460
 461         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 462         if (error)
 463                 return error;
 464
 465         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 466         if (error)
 467                 goto out;
 468         mutex_lock(&sdp->sd_inum_mutex);
 469
 470         error = gfs2_meta_inode_buffer(ip, &bh);
 471         if (error)
 472                 goto out_end_trans;
 473
 474         gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
 475
 476         if (!ir.ir_length) {
 477                 struct buffer_head *m_bh;
 478                 u64 x, y;
 479                 __be64 z;
 480
 481                 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 482                 if (error)
 483                         goto out_brelse;
 484
 485                 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode));
 486                 x = y = be64_to_cpu(z);
 487                 ir.ir_start = x;
 488                 ir.ir_length = GFS2_INUM_QUANTUM;
 489                 x += GFS2_INUM_QUANTUM;
 490                 if (x < y)
 491                         gfs2_consist_inode(m_ip);
 492                 z = cpu_to_be64(x);
 493                 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 494                 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z;
 495
 496                 brelse(m_bh);
 497         }
 498
 499         *formal_ino = ir.ir_start++;
 500         ir.ir_length--;
 501
 502         gfs2_trans_add_bh(ip->i_gl, bh, 1);
 503         gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
 504
 505 out_brelse:
 506         brelse(bh);
 507 out_end_trans:
 508         mutex_unlock(&sdp->sd_inum_mutex);
 509         gfs2_trans_end(sdp);
 510 out:
 511         gfs2_glock_dq_uninit(&gh);
 512         return error;
 513 }
 514
 515 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum)
 516 {
 517         int error;
 518
 519         error = pick_formal_ino_1(sdp, inum);
 520         if (error <= 0)
 521                 return error;
 522
 523         error = pick_formal_ino_2(sdp, inum);
 524
 525         return error;
 526 }
 527
 528 /**
 529  * create_ok - OK to create a new on-disk inode here?
 530  * @dip:  Directory in which dinode is to be created
 531  * @name:  Name of new dinode
 532  * @mode:
 533  *
 534  * Returns: errno
 535  */
 536
 537 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
 538                      unsigned int mode)
 539 {
 540         int error;
 541
 542         error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
 543         if (error)
 544                 return error;
 545
 546         /*  Don't create entries in an unlinked directory  */
 547         if (!dip->i_inode.i_nlink)
 548                 return -EPERM;
 549
 550         error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL);
 551         switch (error) {
 552         case -ENOENT:
 553                 error = 0;
 554                 break;
 555         case 0:
 556                 return -EEXIST;
 557         default:
 558                 return error;
 559         }
 560
 561         if (dip->i_di.di_entries == (u32)-1)
 562                 return -EFBIG;
 563         if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
 564                 return -EMLINK;
 565
 566         return 0;
 567 }
 568
 569 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
 570                                unsigned int *uid, unsigned int *gid)
 571 {
 572         if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
 573             (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
 574                 if (S_ISDIR(*mode))
 575                         *mode |= S_ISUID;
 576                 else if (dip->i_inode.i_uid != current->fsuid)
 577                         *mode &= ~07111;
 578                 *uid = dip->i_inode.i_uid;
 579         } else
 580                 *uid = current->fsuid;
 581
 582         if (dip->i_inode.i_mode & S_ISGID) {
 583                 if (S_ISDIR(*mode))
 584                         *mode |= S_ISGID;
 585                 *gid = dip->i_inode.i_gid;
 586         } else
 587                 *gid = current->fsgid;
 588 }
 589
 590 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum,
 591                         u64 *generation)
 592 {
 593         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 594         int error;
 595
 596         gfs2_alloc_get(dip);
 597
 598         dip->i_alloc.al_requested = RES_DINODE;
 599         error = gfs2_inplace_reserve(dip);
 600         if (error)
 601                 goto out;
 602
 603         error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
 604         if (error)
 605                 goto out_ipreserv;
 606
 607         inum->no_addr = gfs2_alloc_di(dip, generation);
 608
 609         gfs2_trans_end(sdp);
 610
 611 out_ipreserv:
 612         gfs2_inplace_release(dip);
 613 out:
 614         gfs2_alloc_put(dip);
 615         return error;
 616 }
 617
 618 /**
 619  * init_dinode - Fill in a new dinode structure
 620  * @dip: the directory this inode is being created in
 621  * @gl: The glock covering the new inode
 622  * @inum: the inode number
 623  * @mode: the file permissions
 624  * @uid:
 625  * @gid:
 626  *
 627  */
 628
 629 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
 630                         const struct gfs2_inum_host *inum, unsigned int mode,
 631                         unsigned int uid, unsigned int gid,
 632                         const u64 *generation, dev_t dev)
 633 {
 634         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 635         struct gfs2_dinode *di;
 636         struct buffer_head *dibh;
 637
 638         dibh = gfs2_meta_new(gl, inum->no_addr);
 639         gfs2_trans_add_bh(gl, dibh, 1);
 640         gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
 641         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 642         di = (struct gfs2_dinode *)dibh->b_data;
 643
 644         di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
 645         di->di_num.no_addr = cpu_to_be64(inum->no_addr);
 646         di->di_mode = cpu_to_be32(mode);
 647         di->di_uid = cpu_to_be32(uid);
 648         di->di_gid = cpu_to_be32(gid);
 649         di->di_nlink = 0;
 650         di->di_size = 0;
 651         di->di_blocks = cpu_to_be64(1);
 652         di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
 653         di->di_major = cpu_to_be32(MAJOR(dev));
 654         di->di_minor = cpu_to_be32(MINOR(dev));
 655         di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
 656         di->di_generation = cpu_to_be64(*generation);
 657         di->di_flags = 0;
 658
 659         if (S_ISREG(mode)) {
 660                 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
 661                     gfs2_tune_get(sdp, gt_new_files_jdata))
 662                         di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
 663                 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
 664                     gfs2_tune_get(sdp, gt_new_files_directio))
 665                         di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
 666         } else if (S_ISDIR(mode)) {
 667                 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
 668                                             GFS2_DIF_INHERIT_DIRECTIO);
 669                 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
 670                                             GFS2_DIF_INHERIT_JDATA);
 671         }
 672
 673         di->__pad1 = 0;
 674         di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
 675         di->di_height = 0;
 676         di->__pad2 = 0;
 677         di->__pad3 = 0;
 678         di->di_depth = 0;
 679         di->di_entries = 0;
 680         memset(&di->__pad4, 0, sizeof(di->__pad4));
 681         di->di_eattr = 0;
 682         memset(&di->di_reserved, 0, sizeof(di->di_reserved));
 683
 684         brelse(dibh);
 685 }
 686
 687 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
 688                        unsigned int mode, const struct gfs2_inum_host *inum,
 689                        const u64 *generation, dev_t dev)
 690 {
 691         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 692         unsigned int uid, gid;
 693         int error;
 694
 695         munge_mode_uid_gid(dip, &mode, &uid, &gid);
 696         gfs2_alloc_get(dip);
 697
 698         error = gfs2_quota_lock(dip, uid, gid);
 699         if (error)
 700                 goto out;
 701
 702         error = gfs2_quota_check(dip, uid, gid);
 703         if (error)
 704                 goto out_quota;
 705
 706         error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
 707         if (error)
 708                 goto out_quota;
 709
 710         init_dinode(dip, gl, inum, mode, uid, gid, generation, dev);
 711         gfs2_quota_change(dip, +1, uid, gid);
 712         gfs2_trans_end(sdp);
 713
 714 out_quota:
 715         gfs2_quota_unlock(dip);
 716 out:
 717         gfs2_alloc_put(dip);
 718         return error;
 719 }
 720
 721 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
 722                        struct gfs2_inode *ip)
 723 {
 724         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 725         struct gfs2_alloc *al;
 726         int alloc_required;
 727         struct buffer_head *dibh;
 728         int error;
 729
 730         al = gfs2_alloc_get(dip);
 731
 732         error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
 733         if (error)
 734                 goto fail;
 735
 736         error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
 737         if (alloc_required < 0)
 738                 goto fail;
 739         if (alloc_required) {
 740                 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
 741                 if (error)
 742                         goto fail_quota_locks;
 743
 744                 al->al_requested = sdp->sd_max_dirres;
 745
 746                 error = gfs2_inplace_reserve(dip);
 747                 if (error)
 748                         goto fail_quota_locks;
 749
 750                 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
 751                                          al->al_rgd->rd_ri.ri_length +
 752                                          2 * RES_DINODE +
 753                                          RES_STATFS + RES_QUOTA, 0);
 754                 if (error)
 755                         goto fail_ipreserv;
 756         } else {
 757                 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
 758                 if (error)
 759                         goto fail_quota_locks;
 760         }
 761
 762         error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_inode.i_mode));
 763         if (error)
 764                 goto fail_end_trans;
 765
 766         error = gfs2_meta_inode_buffer(ip, &dibh);
 767         if (error)
 768                 goto fail_end_trans;
 769         ip->i_inode.i_nlink = 1;
 770         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 771         gfs2_dinode_out(ip, dibh->b_data);
 772         brelse(dibh);
 773         return 0;
 774
 775 fail_end_trans:
 776         gfs2_trans_end(sdp);
 777
 778 fail_ipreserv:
 779         if (dip->i_alloc.al_rgd)
 780                 gfs2_inplace_release(dip);
 781
 782 fail_quota_locks:
 783         gfs2_quota_unlock(dip);
 784
 785 fail:
 786         gfs2_alloc_put(dip);
 787         return error;
 788 }
 789
 790 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip)
 791 {
 792         int err;
 793         size_t len;
 794         void *value;
 795         char *name;
 796         struct gfs2_ea_request er;
 797
 798         err = security_inode_init_security(&ip->i_inode, &dip->i_inode,
 799                                            &name, &value, &len);
 800
 801         if (err) {
 802                 if (err == -EOPNOTSUPP)
 803                         return 0;
 804                 return err;
 805         }
 806
 807         memset(&er, 0, sizeof(struct gfs2_ea_request));
 808
 809         er.er_type = GFS2_EATYPE_SECURITY;
 810         er.er_name = name;
 811         er.er_data = value;
 812         er.er_name_len = strlen(name);
 813         er.er_data_len = len;
 814
 815         err = gfs2_ea_set_i(ip, &er);
 816
 817         kfree(value);
 818         kfree(name);
 819
 820         return err;
 821 }
 822
 823 /**
 824  * gfs2_createi - Create a new inode
 825  * @ghs: An array of two holders
 826  * @name: The name of the new file
 827  * @mode: the permissions on the new inode
 828  *
 829  * @ghs[0] is an initialized holder for the directory
 830  * @ghs[1] is the holder for the inode lock
 831  *
 832  * If the return value is not NULL, the glocks on both the directory and the new
 833  * file are held.  A transaction has been started and an inplace reservation
 834  * is held, as well.
 835  *
 836  * Returns: An inode
 837  */
 838
 839 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name,
 840                            unsigned int mode, dev_t dev)
 841 {
 842         struct inode *inode;
 843         struct gfs2_inode *dip = ghs->gh_gl->gl_object;
 844         struct inode *dir = &dip->i_inode;
 845         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
 846         struct gfs2_inum_host inum;
 847         int error;
 848         u64 generation;
 849
 850         if (!name->len || name->len > GFS2_FNAMESIZE)
 851                 return ERR_PTR(-ENAMETOOLONG);
 852
 853         gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
 854         error = gfs2_glock_nq(ghs);
 855         if (error)
 856                 goto fail;
 857
 858         error = create_ok(dip, name, mode);
 859         if (error)
 860                 goto fail_gunlock;
 861
 862         error = pick_formal_ino(sdp, &inum.no_formal_ino);
 863         if (error)
 864                 goto fail_gunlock;
 865
 866         error = alloc_dinode(dip, &inum, &generation);
 867         if (error)
 868                 goto fail_gunlock;
 869
 870         error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
 871                                   LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
 872         if (error)
 873                 goto fail_gunlock;
 874
 875         error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev);
 876         if (error)
 877                 goto fail_gunlock2;
 878
 879         inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode));
 880         if (IS_ERR(inode))
 881                 goto fail_gunlock2;
 882
 883         error = gfs2_inode_refresh(GFS2_I(inode));
 884         if (error)
 885                 goto fail_iput;
 886
 887         error = gfs2_acl_create(dip, GFS2_I(inode));
 888         if (error)
 889                 goto fail_iput;
 890
 891         error = gfs2_security_init(dip, GFS2_I(inode));
 892         if (error)
 893                 goto fail_iput;
 894
 895         error = link_dinode(dip, name, GFS2_I(inode));
 896         if (error)
 897                 goto fail_iput;
 898
 899         if (!inode)
 900                 return ERR_PTR(-ENOMEM);
 901         return inode;
 902
 903 fail_iput:
 904         iput(inode);
 905 fail_gunlock2:
 906         gfs2_glock_dq_uninit(ghs + 1);
 907 fail_gunlock:
 908         gfs2_glock_dq(ghs);
 909 fail:
 910         return ERR_PTR(error);
 911 }
 912
 913 /**
 914  * gfs2_rmdiri - Remove a directory
 915  * @dip: The parent directory of the directory to be removed
 916  * @name: The name of the directory to be removed
 917  * @ip: The GFS2 inode of the directory to be removed
 918  *
 919  * Assumes Glocks on dip and ip are held
 920  *
 921  * Returns: errno
 922  */
 923
 924 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name,
 925                 struct gfs2_inode *ip)
 926 {
 927         struct qstr dotname;
 928         int error;
 929
 930         if (ip->i_di.di_entries != 2) {
 931                 if (gfs2_consist_inode(ip))
 932                         gfs2_dinode_print(ip);
 933                 return -EIO;
 934         }
 935
 936         error = gfs2_dir_del(dip, name);
 937         if (error)
 938                 return error;
 939
 940         error = gfs2_change_nlink(dip, -1);
 941         if (error)
 942                 return error;
 943
 944         gfs2_str2qstr(&dotname, ".");
 945         error = gfs2_dir_del(ip, &dotname);
 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         /* It looks odd, but it really should be done twice */
 955         error = gfs2_change_nlink(ip, -1);
 956         if (error)
 957                 return error;
 958
 959         error = gfs2_change_nlink(ip, -1);
 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_host 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_inode.i_mode & S_ISVTX) &&
 988             dip->i_inode.i_uid != current->fsuid &&
 989             ip->i_inode.i_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_inode.i_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         s64 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_inode.i_atime.tv_sec >= 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_inode.i_atime.tv_sec >= quantum) {
1169                         struct buffer_head *dibh;
1170                         struct gfs2_dinode *di;
1171
1172                         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1173                         if (error == -EROFS)
1174                                 return 0;
1175                         if (error)
1176                                 goto fail;
1177
1178                         error = gfs2_meta_inode_buffer(ip, &dibh);
1179                         if (error)
1180                                 goto fail_end_trans;
1181
1182                         ip->i_inode.i_atime.tv_sec = curtime;
1183
1184                         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1185                         di = (struct gfs2_dinode *)dibh->b_data;
1186                         di->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
1187                         brelse(dibh);
1188
1189                         gfs2_trans_end(sdp);
1190                 }
1191
1192                 /* If someone else has asked for the glock,
1193                    unlock and let them have it. Then reacquire
1194                    in the original state. */
1195                 if (gfs2_glock_is_blocking(gl)) {
1196                         gfs2_glock_dq(gh);
1197                         gfs2_holder_reinit(state, flags, gh);
1198                         return gfs2_glock_nq(gh);
1199                 }
1200         }
1201
1202         return 0;
1203
1204 fail_end_trans:
1205         gfs2_trans_end(sdp);
1206 fail:
1207         gfs2_glock_dq(gh);
1208         return error;
1209 }
1210
1211 static int
1212 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1213 {
1214         struct buffer_head *dibh;
1215         int error;
1216
1217         error = gfs2_meta_inode_buffer(ip, &dibh);
1218         if (!error) {
1219                 error = inode_setattr(&ip->i_inode, attr);
1220                 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1221                 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1222                 gfs2_dinode_out(ip, dibh->b_data);
1223                 brelse(dibh);
1224         }
1225         return error;
1226 }
1227
1228 /**
1229  * gfs2_setattr_simple -
1230  * @ip:
1231  * @attr:
1232  *
1233  * Called with a reference on the vnode.
1234  *
1235  * Returns: errno
1236  */
1237
1238 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1239 {
1240         int error;
1241
1242         if (current->journal_info)
1243                 return __gfs2_setattr_simple(ip, attr);
1244
1245         error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
1246         if (error)
1247                 return error;
1248
1249         error = __gfs2_setattr_simple(ip, attr);
1250         gfs2_trans_end(GFS2_SB(&ip->i_inode));
1251         return error;
1252 }
1253