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