fs/orangefs/inode.c

   1 /*
   2  * (C) 2001 Clemson University and The University of Chicago
   3  *
   4  * See COPYING in top-level directory.
   5  */
   6
   7 /*
   8  *  Linux VFS inode operations.
   9  */
  10
  11 #include "protocol.h"
  12 #include "orangefs-kernel.h"
  13 #include "orangefs-bufmap.h"
  14
  15 static int read_one_page(struct page *page)
  16 {
  17         int ret;
  18         int max_block;
  19         ssize_t bytes_read = 0;
  20         struct inode *inode = page->mapping->host;
  21         const __u32 blocksize = PAGE_CACHE_SIZE;        /* inode->i_blksize */
  22         const __u32 blockbits = PAGE_CACHE_SHIFT;       /* inode->i_blkbits */
  23         struct iov_iter to;
  24         struct bio_vec bv = {.bv_page = page, .bv_len = PAGE_SIZE};
  25
  26         iov_iter_bvec(&to, ITER_BVEC | READ, &bv, 1, PAGE_SIZE);
  27
  28         gossip_debug(GOSSIP_INODE_DEBUG,
  29                     "orangefs_readpage called with page %p\n",
  30                      page);
  31
  32         max_block = ((inode->i_size / blocksize) + 1);
  33
  34         if (page->index < max_block) {
  35                 loff_t blockptr_offset = (((loff_t) page->index) << blockbits);
  36
  37                 bytes_read = orangefs_inode_read(inode,
  38                                                  &to,
  39                                                  &blockptr_offset,
  40                                                  inode->i_size);
  41         }
  42         /* this will only zero remaining unread portions of the page data */
  43         iov_iter_zero(~0U, &to);
  44         /* takes care of potential aliasing */
  45         flush_dcache_page(page);
  46         if (bytes_read < 0) {
  47                 ret = bytes_read;
  48                 SetPageError(page);
  49         } else {
  50                 SetPageUptodate(page);
  51                 if (PageError(page))
  52                         ClearPageError(page);
  53                 ret = 0;
  54         }
  55         /* unlock the page after the ->readpage() routine completes */
  56         unlock_page(page);
  57         return ret;
  58 }
  59
  60 static int orangefs_readpage(struct file *file, struct page *page)
  61 {
  62         return read_one_page(page);
  63 }
  64
  65 static int orangefs_readpages(struct file *file,
  66                            struct address_space *mapping,
  67                            struct list_head *pages,
  68                            unsigned nr_pages)
  69 {
  70         int page_idx;
  71         int ret;
  72
  73         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_readpages called\n");
  74
  75         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
  76                 struct page *page;
  77
  78                 page = list_entry(pages->prev, struct page, lru);
  79                 list_del(&page->lru);
  80                 if (!add_to_page_cache(page,
  81                                        mapping,
  82                                        page->index,
  83                                        GFP_KERNEL)) {
  84                         ret = read_one_page(page);
  85                         gossip_debug(GOSSIP_INODE_DEBUG,
  86                                 "failure adding page to cache, read_one_page returned: %d\n",
  87                                 ret);
  88               } else {
  89                         page_cache_release(page);
  90               }
  91         }
  92         BUG_ON(!list_empty(pages));
  93         return 0;
  94 }
  95
  96 static void orangefs_invalidatepage(struct page *page,
  97                                  unsigned int offset,
  98                                  unsigned int length)
  99 {
 100         gossip_debug(GOSSIP_INODE_DEBUG,
 101                      "orangefs_invalidatepage called on page %p "
 102                      "(offset is %u)\n",
 103                      page,
 104                      offset);
 105
 106         ClearPageUptodate(page);
 107         ClearPageMappedToDisk(page);
 108         return;
 109
 110 }
 111
 112 static int orangefs_releasepage(struct page *page, gfp_t foo)
 113 {
 114         gossip_debug(GOSSIP_INODE_DEBUG,
 115                      "orangefs_releasepage called on page %p\n",
 116                      page);
 117         return 0;
 118 }
 119
 120 /*
 121  * Having a direct_IO entry point in the address_space_operations
 122  * struct causes the kernel to allows us to use O_DIRECT on
 123  * open. Nothing will ever call this thing, but in the future we
 124  * will need to be able to use O_DIRECT on open in order to support
 125  * AIO. Modeled after NFS, they do this too.
 126  */
 127 /*
 128  * static ssize_t orangefs_direct_IO(int rw,
 129  *                      struct kiocb *iocb,
 130  *                      struct iov_iter *iter,
 131  *                      loff_t offset)
 132  *{
 133  *      gossip_debug(GOSSIP_INODE_DEBUG,
 134  *                   "orangefs_direct_IO: %s\n",
 135  *                   iocb->ki_filp->f_path.dentry->d_name.name);
 136  *
 137  *      return -EINVAL;
 138  *}
 139  */
 140
 141 struct backing_dev_info orangefs_backing_dev_info = {
 142         .name = "orangefs",
 143         .ra_pages = 0,
 144         .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 145 };
 146
 147 /** ORANGEFS2 implementation of address space operations */
 148 const struct address_space_operations orangefs_address_operations = {
 149         .readpage = orangefs_readpage,
 150         .readpages = orangefs_readpages,
 151         .invalidatepage = orangefs_invalidatepage,
 152         .releasepage = orangefs_releasepage,
 153 /*      .direct_IO = orangefs_direct_IO */
 154 };
 155
 156 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 157 {
 158         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 159         struct orangefs_kernel_op_s *new_op;
 160         loff_t orig_size;
 161         int ret = -EINVAL;
 162
 163         gossip_debug(GOSSIP_INODE_DEBUG,
 164                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 165                      __func__,
 166                      get_khandle_from_ino(inode),
 167                      &orangefs_inode->refn.khandle,
 168                      orangefs_inode->refn.fs_id,
 169                      iattr->ia_size);
 170
 171         /* Ensure that we have a up to date size, so we know if it changed. */
 172         ret = orangefs_inode_getattr(inode, 0, 1);
 173         if (ret == -ESTALE)
 174                 ret = -EIO;
 175         if (ret) {
 176                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 177                     __func__, ret);
 178                 return ret;
 179         }
 180         orig_size = i_size_read(inode);
 181
 182         truncate_setsize(inode, iattr->ia_size);
 183
 184         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 185         if (!new_op)
 186                 return -ENOMEM;
 187
 188         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 189         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 190
 191         ret = service_operation(new_op, __func__,
 192                                 get_interruptible_flag(inode));
 193
 194         /*
 195          * the truncate has no downcall members to retrieve, but
 196          * the status value tells us if it went through ok or not
 197          */
 198         gossip_debug(GOSSIP_INODE_DEBUG,
 199                      "orangefs: orangefs_truncate got return value of %d\n",
 200                      ret);
 201
 202         op_release(new_op);
 203
 204         if (ret != 0)
 205                 return ret;
 206
 207         /*
 208          * Only change the c/mtime if we are changing the size or we are
 209          * explicitly asked to change it.  This handles the semantic difference
 210          * between truncate() and ftruncate() as implemented in the VFS.
 211          *
 212          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
 213          * special case where we need to update the times despite not having
 214          * these flags set.  For all other operations the VFS set these flags
 215          * explicitly if it wants a timestamp update.
 216          */
 217         if (orig_size != i_size_read(inode) &&
 218             !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
 219                 iattr->ia_ctime = iattr->ia_mtime =
 220                         current_fs_time(inode->i_sb);
 221                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 222         }
 223
 224         return ret;
 225 }
 226
 227 /*
 228  * Change attributes of an object referenced by dentry.
 229  */
 230 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
 231 {
 232         int ret = -EINVAL;
 233         struct inode *inode = dentry->d_inode;
 234
 235         gossip_debug(GOSSIP_INODE_DEBUG,
 236                      "orangefs_setattr: called on %s\n",
 237                      dentry->d_name.name);
 238
 239         ret = inode_change_ok(inode, iattr);
 240         if (ret)
 241                 goto out;
 242
 243         if ((iattr->ia_valid & ATTR_SIZE) &&
 244             iattr->ia_size != i_size_read(inode)) {
 245                 ret = orangefs_setattr_size(inode, iattr);
 246                 if (ret)
 247                         goto out;
 248         }
 249
 250         setattr_copy(inode, iattr);
 251         mark_inode_dirty(inode);
 252
 253         ret = orangefs_inode_setattr(inode, iattr);
 254         gossip_debug(GOSSIP_INODE_DEBUG,
 255                      "orangefs_setattr: inode_setattr returned %d\n",
 256                      ret);
 257
 258         if (!ret && (iattr->ia_valid & ATTR_MODE))
 259                 /* change mod on a file that has ACLs */
 260                 ret = posix_acl_chmod(inode, inode->i_mode);
 261
 262 out:
 263         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", ret);
 264         return ret;
 265 }
 266
 267 /*
 268  * Obtain attributes of an object given a dentry
 269  */
 270 int orangefs_getattr(struct vfsmount *mnt,
 271                   struct dentry *dentry,
 272                   struct kstat *kstat)
 273 {
 274         int ret = -ENOENT;
 275         struct inode *inode = dentry->d_inode;
 276         struct orangefs_inode_s *orangefs_inode = NULL;
 277
 278         gossip_debug(GOSSIP_INODE_DEBUG,
 279                      "orangefs_getattr: called on %s\n",
 280                      dentry->d_name.name);
 281
 282         ret = orangefs_inode_getattr(inode, 0, 1);
 283         if (ret == 0) {
 284                 generic_fillattr(inode, kstat);
 285
 286                 /* override block size reported to stat */
 287                 orangefs_inode = ORANGEFS_I(inode);
 288                 kstat->blksize = orangefs_inode->blksize;
 289         }
 290         return ret;
 291 }
 292
 293 int orangefs_permission(struct inode *inode, int mask)
 294 {
 295         int ret;
 296
 297         if (mask & MAY_NOT_BLOCK)
 298                 return -ECHILD;
 299
 300         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 301
 302         /* Make sure the permission (and other common attrs) are up to date. */
 303         ret = orangefs_inode_getattr(inode, 0, 0);
 304         if (ret < 0)
 305                 return ret;
 306
 307         return generic_permission(inode, mask);
 308 }
 309
 310 /* ORANGEDS2 implementation of VFS inode operations for files */
 311 struct inode_operations orangefs_file_inode_operations = {
 312         .get_acl = orangefs_get_acl,
 313         .set_acl = orangefs_set_acl,
 314         .setattr = orangefs_setattr,
 315         .getattr = orangefs_getattr,
 316         .setxattr = generic_setxattr,
 317         .getxattr = generic_getxattr,
 318         .listxattr = orangefs_listxattr,
 319         .removexattr = generic_removexattr,
 320         .permission = orangefs_permission,
 321 };
 322
 323 static int orangefs_init_iops(struct inode *inode)
 324 {
 325         inode->i_mapping->a_ops = &orangefs_address_operations;
 326
 327         switch (inode->i_mode & S_IFMT) {
 328         case S_IFREG:
 329                 inode->i_op = &orangefs_file_inode_operations;
 330                 inode->i_fop = &orangefs_file_operations;
 331                 inode->i_blkbits = PAGE_CACHE_SHIFT;
 332                 break;
 333         case S_IFLNK:
 334                 inode->i_op = &orangefs_symlink_inode_operations;
 335                 break;
 336         case S_IFDIR:
 337                 inode->i_op = &orangefs_dir_inode_operations;
 338                 inode->i_fop = &orangefs_dir_operations;
 339                 break;
 340         default:
 341                 gossip_debug(GOSSIP_INODE_DEBUG,
 342                              "%s: unsupported mode\n",
 343                              __func__);
 344                 return -EINVAL;
 345         }
 346
 347         return 0;
 348 }
 349
 350 /*
 351  * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
 352  * that will be used as a hash-index from where the handle will
 353  * be searched for in the VFS hash table of inodes.
 354  */
 355 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
 356 {
 357         if (!ref)
 358                 return 0;
 359         return orangefs_khandle_to_ino(&(ref->khandle));
 360 }
 361
 362 /*
 363  * Called to set up an inode from iget5_locked.
 364  */
 365 static int orangefs_set_inode(struct inode *inode, void *data)
 366 {
 367         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 368         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
 369         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
 370         return 0;
 371 }
 372
 373 /*
 374  * Called to determine if handles match.
 375  */
 376 static int orangefs_test_inode(struct inode *inode, void *data)
 377 {
 378         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 379         struct orangefs_inode_s *orangefs_inode = NULL;
 380
 381         orangefs_inode = ORANGEFS_I(inode);
 382         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle))
 383                 && orangefs_inode->refn.fs_id == ref->fs_id);
 384 }
 385
 386 /*
 387  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
 388  * file handle.
 389  *
 390  * @sb: the file system super block instance.
 391  * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
 392  */
 393 struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref)
 394 {
 395         struct inode *inode = NULL;
 396         unsigned long hash;
 397         int error;
 398
 399         hash = orangefs_handle_hash(ref);
 400         inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref);
 401         if (!inode || !(inode->i_state & I_NEW))
 402                 return inode;
 403
 404         error = orangefs_inode_getattr(inode, 1, 0);
 405         if (error) {
 406                 iget_failed(inode);
 407                 return ERR_PTR(error);
 408         }
 409
 410         inode->i_ino = hash;    /* needed for stat etc */
 411         orangefs_init_iops(inode);
 412         unlock_new_inode(inode);
 413
 414         gossip_debug(GOSSIP_INODE_DEBUG,
 415                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
 416                      &ref->khandle,
 417                      ref->fs_id,
 418                      hash,
 419                      inode->i_ino);
 420
 421         return inode;
 422 }
 423
 424 /*
 425  * Allocate an inode for a newly created file and insert it into the inode hash.
 426  */
 427 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
 428                 int mode, dev_t dev, struct orangefs_object_kref *ref)
 429 {
 430         unsigned long hash = orangefs_handle_hash(ref);
 431         struct inode *inode;
 432         int error;
 433
 434         gossip_debug(GOSSIP_INODE_DEBUG,
 435                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
 436                      __func__,
 437                      sb,
 438                      MAJOR(dev),
 439                      MINOR(dev),
 440                      mode);
 441
 442         inode = new_inode(sb);
 443         if (!inode)
 444                 return NULL;
 445
 446         orangefs_set_inode(inode, ref);
 447         inode->i_ino = hash;    /* needed for stat etc */
 448
 449         error = orangefs_inode_getattr(inode, 1, 0);
 450         if (error)
 451                 goto out_iput;
 452
 453         orangefs_init_iops(inode);
 454
 455         inode->i_mode = mode;
 456         inode->i_uid = current_fsuid();
 457         inode->i_gid = current_fsgid();
 458         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 459         inode->i_size = PAGE_CACHE_SIZE;
 460         inode->i_rdev = dev;
 461
 462         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
 463         if (error < 0)
 464                 goto out_iput;
 465
 466         gossip_debug(GOSSIP_INODE_DEBUG,
 467                      "Initializing ACL's for inode %pU\n",
 468                      get_khandle_from_ino(inode));
 469         orangefs_init_acl(inode, dir);
 470         return inode;
 471
 472 out_iput:
 473         iput(inode);
 474         return ERR_PTR(error);
 475 }