fs/afs/inode.c

   1 /*
   2  * Copyright (c) 2002 Red Hat, Inc. All rights reserved.
   3  *
   4  * This software may be freely redistributed under the terms of the
   5  * GNU General Public License.
   6  *
   7  * You should have received a copy of the GNU General Public License
   8  * along with this program; if not, write to the Free Software
   9  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  10  *
  11  * Authors: David Woodhouse <dwmw2@infradead.org>
  12  *          David Howells <dhowells@redhat.com>
  13  *
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/module.h>
  18 #include <linux/init.h>
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/sched.h>
  22 #include <linux/mount.h>
  23 #include <linux/namei.h>
  24 #include <linux/iversion.h>
  25 #include "internal.h"
  26
  27 static const struct inode_operations afs_symlink_inode_operations = {
  28         .get_link       = page_get_link,
  29         .listxattr      = afs_listxattr,
  30 };
  31
  32 static noinline void dump_vnode(struct afs_vnode *vnode, struct afs_vnode *parent_vnode)
  33 {
  34         static unsigned long once_only;
  35
  36         pr_warn("kAFS: AFS vnode with undefined type %u\n",
  37                 vnode->status.type);
  38         pr_warn("kAFS: A=%d m=%o s=%llx v=%llx\n",
  39                 vnode->status.abort_code,
  40                 vnode->status.mode,
  41                 vnode->status.size,
  42                 vnode->status.data_version);
  43         pr_warn("kAFS: vnode %llx:%llx:%x\n",
  44                 vnode->fid.vid,
  45                 vnode->fid.vnode,
  46                 vnode->fid.unique);
  47         if (parent_vnode)
  48                 pr_warn("kAFS: dir %llx:%llx:%x\n",
  49                         parent_vnode->fid.vid,
  50                         parent_vnode->fid.vnode,
  51                         parent_vnode->fid.unique);
  52
  53         if (!test_and_set_bit(0, &once_only))
  54                 dump_stack();
  55 }
  56
  57 /*
  58  * Initialise an inode from the vnode status.
  59  */
  60 static int afs_inode_init_from_status(struct afs_vnode *vnode, struct key *key,
  61                                       struct afs_vnode *parent_vnode)
  62 {
  63         struct inode *inode = AFS_VNODE_TO_I(vnode);
  64
  65         _debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
  66                vnode->status.type,
  67                vnode->status.nlink,
  68                (unsigned long long) vnode->status.size,
  69                vnode->status.data_version,
  70                vnode->status.mode);
  71
  72         read_seqlock_excl(&vnode->cb_lock);
  73
  74         afs_update_inode_from_status(vnode, &vnode->status, NULL,
  75                                      AFS_VNODE_NOT_YET_SET);
  76
  77         switch (vnode->status.type) {
  78         case AFS_FTYPE_FILE:
  79                 inode->i_mode   = S_IFREG | vnode->status.mode;
  80                 inode->i_op     = &afs_file_inode_operations;
  81                 inode->i_fop    = &afs_file_operations;
  82                 inode->i_mapping->a_ops = &afs_fs_aops;
  83                 break;
  84         case AFS_FTYPE_DIR:
  85                 inode->i_mode   = S_IFDIR | vnode->status.mode;
  86                 inode->i_op     = &afs_dir_inode_operations;
  87                 inode->i_fop    = &afs_dir_file_operations;
  88                 inode->i_mapping->a_ops = &afs_dir_aops;
  89                 break;
  90         case AFS_FTYPE_SYMLINK:
  91                 /* Symlinks with a mode of 0644 are actually mountpoints. */
  92                 if ((vnode->status.mode & 0777) == 0644) {
  93                         inode->i_flags |= S_AUTOMOUNT;
  94
  95                         set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
  96
  97                         inode->i_mode   = S_IFDIR | 0555;
  98                         inode->i_op     = &afs_mntpt_inode_operations;
  99                         inode->i_fop    = &afs_mntpt_file_operations;
 100                         inode->i_mapping->a_ops = &afs_fs_aops;
 101                 } else {
 102                         inode->i_mode   = S_IFLNK | vnode->status.mode;
 103                         inode->i_op     = &afs_symlink_inode_operations;
 104                         inode->i_mapping->a_ops = &afs_fs_aops;
 105                 }
 106                 inode_nohighmem(inode);
 107                 break;
 108         default:
 109                 dump_vnode(vnode, parent_vnode);
 110                 read_sequnlock_excl(&vnode->cb_lock);
 111                 return afs_protocol_error(NULL, -EBADMSG, afs_eproto_file_type);
 112         }
 113
 114         /*
 115          * Estimate 512 bytes  blocks used, rounded up to nearest 1K
 116          * for consistency with other AFS clients.
 117          */
 118         inode->i_blocks         = ((i_size_read(inode) + 1023) >> 10) << 1;
 119         vnode->invalid_before   = vnode->status.data_version;
 120
 121         read_sequnlock_excl(&vnode->cb_lock);
 122         return 0;
 123 }
 124
 125 /*
 126  * Fetch file status from the volume.
 127  */
 128 int afs_fetch_status(struct afs_vnode *vnode, struct key *key, bool new_inode)
 129 {
 130         struct afs_fs_cursor fc;
 131         int ret;
 132
 133         _enter("%s,{%llx:%llu.%u,S=%lx}",
 134                vnode->volume->name,
 135                vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
 136                vnode->flags);
 137
 138         ret = -ERESTARTSYS;
 139         if (afs_begin_vnode_operation(&fc, vnode, key, true)) {
 140                 while (afs_select_fileserver(&fc)) {
 141                         fc.cb_break = afs_calc_vnode_cb_break(vnode);
 142                         afs_fs_fetch_file_status(&fc, NULL, new_inode);
 143                 }
 144
 145                 afs_check_for_remote_deletion(&fc, fc.vnode);
 146                 afs_vnode_commit_status(&fc, vnode, fc.cb_break);
 147                 ret = afs_end_vnode_operation(&fc);
 148         }
 149
 150         _leave(" = %d", ret);
 151         return ret;
 152 }
 153
 154 /*
 155  * iget5() comparator
 156  */
 157 int afs_iget5_test(struct inode *inode, void *opaque)
 158 {
 159         struct afs_iget_data *data = opaque;
 160         struct afs_vnode *vnode = AFS_FS_I(inode);
 161
 162         return memcmp(&vnode->fid, &data->fid, sizeof(data->fid)) == 0;
 163 }
 164
 165 /*
 166  * iget5() comparator for inode created by autocell operations
 167  *
 168  * These pseudo inodes don't match anything.
 169  */
 170 static int afs_iget5_pseudo_dir_test(struct inode *inode, void *opaque)
 171 {
 172         return 0;
 173 }
 174
 175 /*
 176  * iget5() inode initialiser
 177  */
 178 static int afs_iget5_set(struct inode *inode, void *opaque)
 179 {
 180         struct afs_iget_data *data = opaque;
 181         struct afs_vnode *vnode = AFS_FS_I(inode);
 182
 183         vnode->fid = data->fid;
 184         vnode->volume = data->volume;
 185
 186         /* YFS supports 96-bit vnode IDs, but Linux only supports
 187          * 64-bit inode numbers.
 188          */
 189         inode->i_ino = data->fid.vnode;
 190         inode->i_generation = data->fid.unique;
 191         return 0;
 192 }
 193
 194 /*
 195  * Create an inode for a dynamic root directory or an autocell dynamic
 196  * automount dir.
 197  */
 198 struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
 199 {
 200         struct afs_iget_data data;
 201         struct afs_super_info *as;
 202         struct afs_vnode *vnode;
 203         struct inode *inode;
 204         static atomic_t afs_autocell_ino;
 205
 206         _enter("");
 207
 208         as = sb->s_fs_info;
 209         if (as->volume) {
 210                 data.volume = as->volume;
 211                 data.fid.vid = as->volume->vid;
 212         }
 213         if (root) {
 214                 data.fid.vnode = 1;
 215                 data.fid.unique = 1;
 216         } else {
 217                 data.fid.vnode = atomic_inc_return(&afs_autocell_ino);
 218                 data.fid.unique = 0;
 219         }
 220
 221         inode = iget5_locked(sb, data.fid.vnode,
 222                              afs_iget5_pseudo_dir_test, afs_iget5_set,
 223                              &data);
 224         if (!inode) {
 225                 _leave(" = -ENOMEM");
 226                 return ERR_PTR(-ENOMEM);
 227         }
 228
 229         _debug("GOT INODE %p { ino=%lu, vl=%llx, vn=%llx, u=%x }",
 230                inode, inode->i_ino, data.fid.vid, data.fid.vnode,
 231                data.fid.unique);
 232
 233         vnode = AFS_FS_I(inode);
 234
 235         /* there shouldn't be an existing inode */
 236         BUG_ON(!(inode->i_state & I_NEW));
 237
 238         inode->i_size           = 0;
 239         inode->i_mode           = S_IFDIR | S_IRUGO | S_IXUGO;
 240         if (root) {
 241                 inode->i_op     = &afs_dynroot_inode_operations;
 242                 inode->i_fop    = &afs_dynroot_file_operations;
 243         } else {
 244                 inode->i_op     = &afs_autocell_inode_operations;
 245         }
 246         set_nlink(inode, 2);
 247         inode->i_uid            = GLOBAL_ROOT_UID;
 248         inode->i_gid            = GLOBAL_ROOT_GID;
 249         inode->i_ctime = inode->i_atime = inode->i_mtime = current_time(inode);
 250         inode->i_blocks         = 0;
 251         inode_set_iversion_raw(inode, 0);
 252         inode->i_generation     = 0;
 253
 254         set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
 255         if (!root) {
 256                 set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
 257                 inode->i_flags |= S_AUTOMOUNT;
 258         }
 259
 260         inode->i_flags |= S_NOATIME;
 261         unlock_new_inode(inode);
 262         _leave(" = %p", inode);
 263         return inode;
 264 }
 265
 266 /*
 267  * Get a cache cookie for an inode.
 268  */
 269 static void afs_get_inode_cache(struct afs_vnode *vnode)
 270 {
 271 #ifdef CONFIG_AFS_FSCACHE
 272         struct {
 273                 u32 vnode_id;
 274                 u32 unique;
 275                 u32 vnode_id_ext[2];    /* Allow for a 96-bit key */
 276         } __packed key;
 277         struct afs_vnode_cache_aux aux;
 278
 279         if (vnode->status.type == AFS_FTYPE_DIR) {
 280                 vnode->cache = NULL;
 281                 return;
 282         }
 283
 284         key.vnode_id            = vnode->fid.vnode;
 285         key.unique              = vnode->fid.unique;
 286         key.vnode_id_ext[0]     = vnode->fid.vnode >> 32;
 287         key.vnode_id_ext[1]     = vnode->fid.vnode_hi;
 288         aux.data_version        = vnode->status.data_version;
 289
 290         vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
 291                                               &afs_vnode_cache_index_def,
 292                                               &key, sizeof(key),
 293                                               &aux, sizeof(aux),
 294                                               vnode, vnode->status.size, true);
 295 #endif
 296 }
 297
 298 /*
 299  * inode retrieval
 300  */
 301 struct inode *afs_iget(struct super_block *sb, struct key *key,
 302                        struct afs_fid *fid, struct afs_file_status *status,
 303                        struct afs_callback *cb, struct afs_cb_interest *cbi,
 304                        struct afs_vnode *parent_vnode)
 305 {
 306         struct afs_iget_data data = { .fid = *fid };
 307         struct afs_super_info *as;
 308         struct afs_vnode *vnode;
 309         struct inode *inode;
 310         int ret;
 311
 312         _enter(",{%llx:%llu.%u},,", fid->vid, fid->vnode, fid->unique);
 313
 314         as = sb->s_fs_info;
 315         data.volume = as->volume;
 316
 317         inode = iget5_locked(sb, fid->vnode, afs_iget5_test, afs_iget5_set,
 318                              &data);
 319         if (!inode) {
 320                 _leave(" = -ENOMEM");
 321                 return ERR_PTR(-ENOMEM);
 322         }
 323
 324         _debug("GOT INODE %p { vl=%llx vn=%llx, u=%x }",
 325                inode, fid->vid, fid->vnode, fid->unique);
 326
 327         vnode = AFS_FS_I(inode);
 328
 329         /* deal with an existing inode */
 330         if (!(inode->i_state & I_NEW)) {
 331                 _leave(" = %p", inode);
 332                 return inode;
 333         }
 334
 335         if (!status) {
 336                 /* it's a remotely extant inode */
 337                 ret = afs_fetch_status(vnode, key, true);
 338                 if (ret < 0)
 339                         goto bad_inode;
 340         } else {
 341                 /* it's an inode we just created */
 342                 memcpy(&vnode->status, status, sizeof(vnode->status));
 343
 344                 if (!cb) {
 345                         /* it's a symlink we just created (the fileserver
 346                          * didn't give us a callback) */
 347                         vnode->cb_version = 0;
 348                         vnode->cb_type = 0;
 349                         vnode->cb_expires_at = ktime_get();
 350                 } else {
 351                         vnode->cb_version = cb->version;
 352                         vnode->cb_type = cb->type;
 353                         vnode->cb_expires_at = cb->expires_at;
 354                         vnode->cb_interest = afs_get_cb_interest(cbi);
 355                         set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
 356                 }
 357
 358                 vnode->cb_expires_at += ktime_get_real_seconds();
 359         }
 360
 361         ret = afs_inode_init_from_status(vnode, key, parent_vnode);
 362         if (ret < 0)
 363                 goto bad_inode;
 364
 365         afs_get_inode_cache(vnode);
 366
 367         /* success */
 368         clear_bit(AFS_VNODE_UNSET, &vnode->flags);
 369         inode->i_flags |= S_NOATIME;
 370         unlock_new_inode(inode);
 371         _leave(" = %p [CB { v=%u t=%u }]", inode, vnode->cb_version, vnode->cb_type);
 372         return inode;
 373
 374         /* failure */
 375 bad_inode:
 376         iget_failed(inode);
 377         _leave(" = %d [bad]", ret);
 378         return ERR_PTR(ret);
 379 }
 380
 381 /*
 382  * mark the data attached to an inode as obsolete due to a write on the server
 383  * - might also want to ditch all the outstanding writes and dirty pages
 384  */
 385 void afs_zap_data(struct afs_vnode *vnode)
 386 {
 387         _enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
 388
 389 #ifdef CONFIG_AFS_FSCACHE
 390         fscache_invalidate(vnode->cache);
 391 #endif
 392
 393         /* nuke all the non-dirty pages that aren't locked, mapped or being
 394          * written back in a regular file and completely discard the pages in a
 395          * directory or symlink */
 396         if (S_ISREG(vnode->vfs_inode.i_mode))
 397                 invalidate_remote_inode(&vnode->vfs_inode);
 398         else
 399                 invalidate_inode_pages2(vnode->vfs_inode.i_mapping);
 400 }
 401
 402 /*
 403  * validate a vnode/inode
 404  * - there are several things we need to check
 405  *   - parent dir data changes (rm, rmdir, rename, mkdir, create, link,
 406  *     symlink)
 407  *   - parent dir metadata changed (security changes)
 408  *   - dentry data changed (write, truncate)
 409  *   - dentry metadata changed (security changes)
 410  */
 411 int afs_validate(struct afs_vnode *vnode, struct key *key)
 412 {
 413         time64_t now = ktime_get_real_seconds();
 414         bool valid;
 415         int ret;
 416
 417         _enter("{v={%llx:%llu} fl=%lx},%x",
 418                vnode->fid.vid, vnode->fid.vnode, vnode->flags,
 419                key_serial(key));
 420
 421         /* Quickly check the callback state.  Ideally, we'd use read_seqbegin
 422          * here, but we have no way to pass the net namespace to the RCU
 423          * cleanup for the server record.
 424          */
 425         read_seqlock_excl(&vnode->cb_lock);
 426
 427         if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
 428                 if (vnode->cb_s_break != vnode->cb_interest->server->cb_s_break ||
 429                     vnode->cb_v_break != vnode->volume->cb_v_break) {
 430                         vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
 431                         vnode->cb_v_break = vnode->volume->cb_v_break;
 432                         valid = false;
 433                 } else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) {
 434                         valid = false;
 435                 } else if (vnode->cb_expires_at - 10 <= now) {
 436                         valid = false;
 437                 } else {
 438                         valid = true;
 439                 }
 440         } else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
 441                 valid = true;
 442         } else {
 443                 vnode->cb_v_break = vnode->volume->cb_v_break;
 444                 valid = false;
 445         }
 446
 447         read_sequnlock_excl(&vnode->cb_lock);
 448
 449         if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
 450                 clear_nlink(&vnode->vfs_inode);
 451
 452         if (valid)
 453                 goto valid;
 454
 455         down_write(&vnode->validate_lock);
 456
 457         /* if the promise has expired, we need to check the server again to get
 458          * a new promise - note that if the (parent) directory's metadata was
 459          * changed then the security may be different and we may no longer have
 460          * access */
 461         if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
 462                 _debug("not promised");
 463                 ret = afs_fetch_status(vnode, key, false);
 464                 if (ret < 0) {
 465                         if (ret == -ENOENT) {
 466                                 set_bit(AFS_VNODE_DELETED, &vnode->flags);
 467                                 ret = -ESTALE;
 468                         }
 469                         goto error_unlock;
 470                 }
 471                 _debug("new promise [fl=%lx]", vnode->flags);
 472         }
 473
 474         if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
 475                 _debug("file already deleted");
 476                 ret = -ESTALE;
 477                 goto error_unlock;
 478         }
 479
 480         /* if the vnode's data version number changed then its contents are
 481          * different */
 482         if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
 483                 afs_zap_data(vnode);
 484         up_write(&vnode->validate_lock);
 485 valid:
 486         _leave(" = 0");
 487         return 0;
 488
 489 error_unlock:
 490         up_write(&vnode->validate_lock);
 491         _leave(" = %d", ret);
 492         return ret;
 493 }
 494
 495 /*
 496  * read the attributes of an inode
 497  */
 498 int afs_getattr(const struct path *path, struct kstat *stat,
 499                 u32 request_mask, unsigned int query_flags)
 500 {
 501         struct inode *inode = d_inode(path->dentry);
 502         struct afs_vnode *vnode = AFS_FS_I(inode);
 503         int seq = 0;
 504
 505         _enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
 506
 507         do {
 508                 read_seqbegin_or_lock(&vnode->cb_lock, &seq);
 509                 generic_fillattr(inode, stat);
 510         } while (need_seqretry(&vnode->cb_lock, seq));
 511
 512         done_seqretry(&vnode->cb_lock, seq);
 513         return 0;
 514 }
 515
 516 /*
 517  * discard an AFS inode
 518  */
 519 int afs_drop_inode(struct inode *inode)
 520 {
 521         _enter("");
 522
 523         if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
 524                 return generic_delete_inode(inode);
 525         else
 526                 return generic_drop_inode(inode);
 527 }
 528
 529 /*
 530  * clear an AFS inode
 531  */
 532 void afs_evict_inode(struct inode *inode)
 533 {
 534         struct afs_vnode *vnode;
 535
 536         vnode = AFS_FS_I(inode);
 537
 538         _enter("{%llx:%llu.%d}",
 539                vnode->fid.vid,
 540                vnode->fid.vnode,
 541                vnode->fid.unique);
 542
 543         _debug("CLEAR INODE %p", inode);
 544
 545         ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
 546
 547         truncate_inode_pages_final(&inode->i_data);
 548         clear_inode(inode);
 549
 550         if (vnode->cb_interest) {
 551                 afs_put_cb_interest(afs_i2net(inode), vnode->cb_interest);
 552                 vnode->cb_interest = NULL;
 553         }
 554
 555         while (!list_empty(&vnode->wb_keys)) {
 556                 struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
 557                                                     struct afs_wb_key, vnode_link);
 558                 list_del(&wbk->vnode_link);
 559                 afs_put_wb_key(wbk);
 560         }
 561
 562 #ifdef CONFIG_AFS_FSCACHE
 563         {
 564                 struct afs_vnode_cache_aux aux;
 565
 566                 aux.data_version = vnode->status.data_version;
 567                 fscache_relinquish_cookie(vnode->cache, &aux,
 568                                           test_bit(AFS_VNODE_DELETED, &vnode->flags));
 569                 vnode->cache = NULL;
 570         }
 571 #endif
 572
 573         afs_prune_wb_keys(vnode);
 574         afs_put_permits(rcu_access_pointer(vnode->permit_cache));
 575         key_put(vnode->silly_key);
 576         vnode->silly_key = NULL;
 577         key_put(vnode->lock_key);
 578         vnode->lock_key = NULL;
 579         _leave("");
 580 }
 581
 582 /*
 583  * set the attributes of an inode
 584  */
 585 int afs_setattr(struct dentry *dentry, struct iattr *attr)
 586 {
 587         struct afs_fs_cursor fc;
 588         struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
 589         struct key *key;
 590         int ret;
 591
 592         _enter("{%llx:%llu},{n=%pd},%x",
 593                vnode->fid.vid, vnode->fid.vnode, dentry,
 594                attr->ia_valid);
 595
 596         if (!(attr->ia_valid & (ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
 597                                 ATTR_MTIME))) {
 598                 _leave(" = 0 [unsupported]");
 599                 return 0;
 600         }
 601
 602         /* flush any dirty data outstanding on a regular file */
 603         if (S_ISREG(vnode->vfs_inode.i_mode))
 604                 filemap_write_and_wait(vnode->vfs_inode.i_mapping);
 605
 606         if (attr->ia_valid & ATTR_FILE) {
 607                 key = afs_file_key(attr->ia_file);
 608         } else {
 609                 key = afs_request_key(vnode->volume->cell);
 610                 if (IS_ERR(key)) {
 611                         ret = PTR_ERR(key);
 612                         goto error;
 613                 }
 614         }
 615
 616         ret = -ERESTARTSYS;
 617         if (afs_begin_vnode_operation(&fc, vnode, key, false)) {
 618                 while (afs_select_fileserver(&fc)) {
 619                         fc.cb_break = afs_calc_vnode_cb_break(vnode);
 620                         afs_fs_setattr(&fc, attr);
 621                 }
 622
 623                 afs_check_for_remote_deletion(&fc, fc.vnode);
 624                 afs_vnode_commit_status(&fc, vnode, fc.cb_break);
 625                 ret = afs_end_vnode_operation(&fc);
 626         }
 627
 628         if (!(attr->ia_valid & ATTR_FILE))
 629                 key_put(key);
 630
 631 error:
 632         _leave(" = %d", ret);
 633         return ret;
 634 }