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 struct backing_dev_info orangefs_backing_dev_info = {
 140         .name = "orangefs",
 141         .ra_pages = 0,
 142         .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 143 };
 144
 145 /** ORANGEFS2 implementation of address space operations */
 146 const struct address_space_operations orangefs_address_operations = {
 147         .readpage = orangefs_readpage,
 148         .readpages = orangefs_readpages,
 149         .invalidatepage = orangefs_invalidatepage,
 150         .releasepage = orangefs_releasepage,
 151         .direct_IO = orangefs_direct_IO,
 152 };
 153
 154 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 155 {
 156         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 157         struct orangefs_kernel_op_s *new_op;
 158         loff_t orig_size;
 159         int ret = -EINVAL;
 160
 161         gossip_debug(GOSSIP_INODE_DEBUG,
 162                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 163                      __func__,
 164                      get_khandle_from_ino(inode),
 165                      &orangefs_inode->refn.khandle,
 166                      orangefs_inode->refn.fs_id,
 167                      iattr->ia_size);
 168
 169         /* Ensure that we have a up to date size, so we know if it changed. */
 170         ret = orangefs_inode_getattr(inode, 0, 1);
 171         if (ret == -ESTALE)
 172                 ret = -EIO;
 173         if (ret) {
 174                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 175                     __func__, ret);
 176                 return ret;
 177         }
 178         orig_size = i_size_read(inode);
 179
 180         truncate_setsize(inode, iattr->ia_size);
 181
 182         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 183         if (!new_op)
 184                 return -ENOMEM;
 185
 186         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 187         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 188
 189         ret = service_operation(new_op, __func__,
 190                                 get_interruptible_flag(inode));
 191
 192         /*
 193          * the truncate has no downcall members to retrieve, but
 194          * the status value tells us if it went through ok or not
 195          */
 196         gossip_debug(GOSSIP_INODE_DEBUG,
 197                      "orangefs: orangefs_truncate got return value of %d\n",
 198                      ret);
 199
 200         op_release(new_op);
 201
 202         if (ret != 0)
 203                 return ret;
 204
 205         if (orig_size != i_size_read(inode))
 206                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 207
 208         return ret;
 209 }
 210
 211 /*
 212  * Change attributes of an object referenced by dentry.
 213  */
 214 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
 215 {
 216         int ret = -EINVAL;
 217         struct inode *inode = dentry->d_inode;
 218
 219         gossip_debug(GOSSIP_INODE_DEBUG,
 220                      "orangefs_setattr: called on %pd\n",
 221                      dentry);
 222
 223         ret = setattr_prepare(dentry, iattr);
 224         if (ret)
 225                 goto out;
 226
 227         if (iattr->ia_valid & ATTR_SIZE) {
 228                 ret = orangefs_setattr_size(inode, iattr);
 229                 if (ret)
 230                         goto out;
 231         }
 232
 233         setattr_copy(inode, iattr);
 234         mark_inode_dirty(inode);
 235
 236         ret = orangefs_inode_setattr(inode, iattr);
 237         gossip_debug(GOSSIP_INODE_DEBUG,
 238                      "orangefs_setattr: inode_setattr returned %d\n",
 239                      ret);
 240
 241         if (!ret && (iattr->ia_valid & ATTR_MODE))
 242                 /* change mod on a file that has ACLs */
 243                 ret = posix_acl_chmod(inode, inode->i_mode);
 244
 245 out:
 246         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", ret);
 247         return ret;
 248 }
 249
 250 /*
 251  * Obtain attributes of an object given a dentry
 252  */
 253 int orangefs_getattr(struct vfsmount *mnt,
 254                   struct dentry *dentry,
 255                   struct kstat *kstat)
 256 {
 257         int ret = -ENOENT;
 258         struct inode *inode = dentry->d_inode;
 259         struct orangefs_inode_s *orangefs_inode = NULL;
 260
 261         gossip_debug(GOSSIP_INODE_DEBUG,
 262                      "orangefs_getattr: called on %pd\n",
 263                      dentry);
 264
 265         ret = orangefs_inode_getattr(inode, 0, 0);
 266         if (ret == 0) {
 267                 generic_fillattr(inode, kstat);
 268
 269                 /* override block size reported to stat */
 270                 orangefs_inode = ORANGEFS_I(inode);
 271                 kstat->blksize = orangefs_inode->blksize;
 272         }
 273         return ret;
 274 }
 275
 276 int orangefs_permission(struct inode *inode, int mask)
 277 {
 278         int ret;
 279
 280         if (mask & MAY_NOT_BLOCK)
 281                 return -ECHILD;
 282
 283         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 284
 285         /* Make sure the permission (and other common attrs) are up to date. */
 286         ret = orangefs_inode_getattr(inode, 0, 0);
 287         if (ret < 0)
 288                 return ret;
 289
 290         return generic_permission(inode, mask);
 291 }
 292
 293 /* ORANGEDS2 implementation of VFS inode operations for files */
 294 const struct inode_operations orangefs_file_inode_operations = {
 295         .get_acl = orangefs_get_acl,
 296         .set_acl = orangefs_set_acl,
 297         .setattr = orangefs_setattr,
 298         .getattr = orangefs_getattr,
 299         .listxattr = orangefs_listxattr,
 300         .permission = orangefs_permission,
 301 };
 302
 303 static int orangefs_init_iops(struct inode *inode)
 304 {
 305         inode->i_mapping->a_ops = &orangefs_address_operations;
 306
 307         switch (inode->i_mode & S_IFMT) {
 308         case S_IFREG:
 309                 inode->i_op = &orangefs_file_inode_operations;
 310                 inode->i_fop = &orangefs_file_operations;
 311                 inode->i_blkbits = PAGE_SHIFT;
 312                 break;
 313         case S_IFLNK:
 314                 inode->i_op = &orangefs_symlink_inode_operations;
 315                 break;
 316         case S_IFDIR:
 317                 inode->i_op = &orangefs_dir_inode_operations;
 318                 inode->i_fop = &orangefs_dir_operations;
 319                 break;
 320         default:
 321                 gossip_debug(GOSSIP_INODE_DEBUG,
 322                              "%s: unsupported mode\n",
 323                              __func__);
 324                 return -EINVAL;
 325         }
 326
 327         return 0;
 328 }
 329
 330 /*
 331  * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
 332  * that will be used as a hash-index from where the handle will
 333  * be searched for in the VFS hash table of inodes.
 334  */
 335 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
 336 {
 337         if (!ref)
 338                 return 0;
 339         return orangefs_khandle_to_ino(&(ref->khandle));
 340 }
 341
 342 /*
 343  * Called to set up an inode from iget5_locked.
 344  */
 345 static int orangefs_set_inode(struct inode *inode, void *data)
 346 {
 347         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 348         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
 349         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
 350         return 0;
 351 }
 352
 353 /*
 354  * Called to determine if handles match.
 355  */
 356 static int orangefs_test_inode(struct inode *inode, void *data)
 357 {
 358         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 359         struct orangefs_inode_s *orangefs_inode = NULL;
 360
 361         orangefs_inode = ORANGEFS_I(inode);
 362         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle))
 363                 && orangefs_inode->refn.fs_id == ref->fs_id);
 364 }
 365
 366 /*
 367  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
 368  * file handle.
 369  *
 370  * @sb: the file system super block instance.
 371  * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
 372  */
 373 struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref)
 374 {
 375         struct inode *inode = NULL;
 376         unsigned long hash;
 377         int error;
 378
 379         hash = orangefs_handle_hash(ref);
 380         inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref);
 381         if (!inode || !(inode->i_state & I_NEW))
 382                 return inode;
 383
 384         error = orangefs_inode_getattr(inode, 1, 1);
 385         if (error) {
 386                 iget_failed(inode);
 387                 return ERR_PTR(error);
 388         }
 389
 390         inode->i_ino = hash;    /* needed for stat etc */
 391         orangefs_init_iops(inode);
 392         unlock_new_inode(inode);
 393
 394         gossip_debug(GOSSIP_INODE_DEBUG,
 395                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
 396                      &ref->khandle,
 397                      ref->fs_id,
 398                      hash,
 399                      inode->i_ino);
 400
 401         return inode;
 402 }
 403
 404 /*
 405  * Allocate an inode for a newly created file and insert it into the inode hash.
 406  */
 407 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
 408                 int mode, dev_t dev, struct orangefs_object_kref *ref)
 409 {
 410         unsigned long hash = orangefs_handle_hash(ref);
 411         struct inode *inode;
 412         int error;
 413
 414         gossip_debug(GOSSIP_INODE_DEBUG,
 415                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
 416                      __func__,
 417                      sb,
 418                      MAJOR(dev),
 419                      MINOR(dev),
 420                      mode);
 421
 422         inode = new_inode(sb);
 423         if (!inode)
 424                 return NULL;
 425
 426         orangefs_set_inode(inode, ref);
 427         inode->i_ino = hash;    /* needed for stat etc */
 428
 429         error = orangefs_inode_getattr(inode, 1, 1);
 430         if (error)
 431                 goto out_iput;
 432
 433         orangefs_init_iops(inode);
 434
 435         inode->i_mode = mode;
 436         inode->i_uid = current_fsuid();
 437         inode->i_gid = current_fsgid();
 438         inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 439         inode->i_size = PAGE_SIZE;
 440         inode->i_rdev = dev;
 441
 442         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
 443         if (error < 0)
 444                 goto out_iput;
 445
 446         gossip_debug(GOSSIP_INODE_DEBUG,
 447                      "Initializing ACL's for inode %pU\n",
 448                      get_khandle_from_ino(inode));
 449         orangefs_init_acl(inode, dir);
 450         return inode;
 451
 452 out_iput:
 453         iput(inode);
 454         return ERR_PTR(error);
 455 }