fs/orangefs/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * (C) 2001 Clemson University and The University of Chicago
   4  *
   5  * See COPYING in top-level directory.
   6  */
   7
   8 /*
   9  *  Linux VFS inode operations.
  10  */
  11
  12 #include <linux/bvec.h>
  13 #include "protocol.h"
  14 #include "orangefs-kernel.h"
  15 #include "orangefs-bufmap.h"
  16
  17 static int read_one_page(struct page *page)
  18 {
  19         int ret;
  20         int max_block;
  21         ssize_t bytes_read = 0;
  22         struct inode *inode = page->mapping->host;
  23         const __u32 blocksize = PAGE_SIZE;      /* inode->i_blksize */
  24         const __u32 blockbits = PAGE_SHIFT;     /* inode->i_blkbits */
  25         struct iov_iter to;
  26         struct bio_vec bv = {.bv_page = page, .bv_len = PAGE_SIZE};
  27
  28         iov_iter_bvec(&to, ITER_BVEC | READ, &bv, 1, PAGE_SIZE);
  29
  30         gossip_debug(GOSSIP_INODE_DEBUG,
  31                     "orangefs_readpage called with page %p\n",
  32                      page);
  33
  34         max_block = ((inode->i_size / blocksize) + 1);
  35
  36         if (page->index < max_block) {
  37                 loff_t blockptr_offset = (((loff_t) page->index) << blockbits);
  38
  39                 bytes_read = orangefs_inode_read(inode,
  40                                                  &to,
  41                                                  &blockptr_offset,
  42                                                  inode->i_size);
  43         }
  44         /* this will only zero remaining unread portions of the page data */
  45         iov_iter_zero(~0U, &to);
  46         /* takes care of potential aliasing */
  47         flush_dcache_page(page);
  48         if (bytes_read < 0) {
  49                 ret = bytes_read;
  50                 SetPageError(page);
  51         } else {
  52                 SetPageUptodate(page);
  53                 if (PageError(page))
  54                         ClearPageError(page);
  55                 ret = 0;
  56         }
  57         /* unlock the page after the ->readpage() routine completes */
  58         unlock_page(page);
  59         return ret;
  60 }
  61
  62 static int orangefs_readpage(struct file *file, struct page *page)
  63 {
  64         return read_one_page(page);
  65 }
  66
  67 static int orangefs_readpages(struct file *file,
  68                            struct address_space *mapping,
  69                            struct list_head *pages,
  70                            unsigned nr_pages)
  71 {
  72         int page_idx;
  73         int ret;
  74
  75         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_readpages called\n");
  76
  77         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
  78                 struct page *page;
  79
  80                 page = list_entry(pages->prev, struct page, lru);
  81                 list_del(&page->lru);
  82                 if (!add_to_page_cache(page,
  83                                        mapping,
  84                                        page->index,
  85                                        readahead_gfp_mask(mapping))) {
  86                         ret = read_one_page(page);
  87                         gossip_debug(GOSSIP_INODE_DEBUG,
  88                                 "failure adding page to cache, read_one_page returned: %d\n",
  89                                 ret);
  90               } else {
  91                         put_page(page);
  92               }
  93         }
  94         BUG_ON(!list_empty(pages));
  95         return 0;
  96 }
  97
  98 static void orangefs_invalidatepage(struct page *page,
  99                                  unsigned int offset,
 100                                  unsigned int length)
 101 {
 102         gossip_debug(GOSSIP_INODE_DEBUG,
 103                      "orangefs_invalidatepage called on page %p "
 104                      "(offset is %u)\n",
 105                      page,
 106                      offset);
 107
 108         ClearPageUptodate(page);
 109         ClearPageMappedToDisk(page);
 110         return;
 111
 112 }
 113
 114 static int orangefs_releasepage(struct page *page, gfp_t foo)
 115 {
 116         gossip_debug(GOSSIP_INODE_DEBUG,
 117                      "orangefs_releasepage called on page %p\n",
 118                      page);
 119         return 0;
 120 }
 121
 122 /*
 123  * Having a direct_IO entry point in the address_space_operations
 124  * struct causes the kernel to allows us to use O_DIRECT on
 125  * open. Nothing will ever call this thing, but in the future we
 126  * will need to be able to use O_DIRECT on open in order to support
 127  * AIO. Modeled after NFS, they do this too.
 128  */
 129
 130 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
 131                                   struct iov_iter *iter)
 132 {
 133         gossip_debug(GOSSIP_INODE_DEBUG,
 134                      "orangefs_direct_IO: %pD\n",
 135                      iocb->ki_filp);
 136
 137         return -EINVAL;
 138 }
 139
 140 /** ORANGEFS2 implementation of address space operations */
 141 const struct address_space_operations orangefs_address_operations = {
 142         .readpage = orangefs_readpage,
 143         .readpages = orangefs_readpages,
 144         .invalidatepage = orangefs_invalidatepage,
 145         .releasepage = orangefs_releasepage,
 146         .direct_IO = orangefs_direct_IO,
 147 };
 148
 149 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 150 {
 151         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 152         struct orangefs_kernel_op_s *new_op;
 153         loff_t orig_size;
 154         int ret = -EINVAL;
 155
 156         gossip_debug(GOSSIP_INODE_DEBUG,
 157                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 158                      __func__,
 159                      get_khandle_from_ino(inode),
 160                      &orangefs_inode->refn.khandle,
 161                      orangefs_inode->refn.fs_id,
 162                      iattr->ia_size);
 163
 164         /* Ensure that we have a up to date size, so we know if it changed. */
 165         ret = orangefs_inode_getattr(inode, 0, 1, STATX_SIZE);
 166         if (ret == -ESTALE)
 167                 ret = -EIO;
 168         if (ret) {
 169                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 170                     __func__, ret);
 171                 return ret;
 172         }
 173         orig_size = i_size_read(inode);
 174
 175         truncate_setsize(inode, iattr->ia_size);
 176
 177         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 178         if (!new_op)
 179                 return -ENOMEM;
 180
 181         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 182         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 183
 184         ret = service_operation(new_op, __func__,
 185                                 get_interruptible_flag(inode));
 186
 187         /*
 188          * the truncate has no downcall members to retrieve, but
 189          * the status value tells us if it went through ok or not
 190          */
 191         gossip_debug(GOSSIP_INODE_DEBUG,
 192                      "orangefs: orangefs_truncate got return value of %d\n",
 193                      ret);
 194
 195         op_release(new_op);
 196
 197         if (ret != 0)
 198                 return ret;
 199
 200         if (orig_size != i_size_read(inode))
 201                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 202
 203         return ret;
 204 }
 205
 206 /*
 207  * Change attributes of an object referenced by dentry.
 208  */
 209 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
 210 {
 211         int ret = -EINVAL;
 212         struct inode *inode = dentry->d_inode;
 213
 214         gossip_debug(GOSSIP_INODE_DEBUG,
 215                      "orangefs_setattr: called on %pd\n",
 216                      dentry);
 217
 218         ret = setattr_prepare(dentry, iattr);
 219         if (ret)
 220                 goto out;
 221
 222         if (iattr->ia_valid & ATTR_SIZE) {
 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(const struct path *path, struct kstat *stat,
 249                      u32 request_mask, unsigned int flags)
 250 {
 251         int ret = -ENOENT;
 252         struct inode *inode = path->dentry->d_inode;
 253         struct orangefs_inode_s *orangefs_inode = NULL;
 254
 255         gossip_debug(GOSSIP_INODE_DEBUG,
 256                      "orangefs_getattr: called on %pd\n",
 257                      path->dentry);
 258
 259         ret = orangefs_inode_getattr(inode, 0, 0, request_mask);
 260         if (ret == 0) {
 261                 generic_fillattr(inode, stat);
 262
 263                 /* override block size reported to stat */
 264                 orangefs_inode = ORANGEFS_I(inode);
 265                 stat->blksize = orangefs_inode->blksize;
 266
 267                 if (request_mask & STATX_SIZE)
 268                         stat->result_mask = STATX_BASIC_STATS;
 269                 else
 270                         stat->result_mask = STATX_BASIC_STATS &
 271                             ~STATX_SIZE;
 272
 273                 stat->attributes_mask = STATX_ATTR_IMMUTABLE |
 274                     STATX_ATTR_APPEND;
 275                 if (inode->i_flags & S_IMMUTABLE)
 276                         stat->attributes |= STATX_ATTR_IMMUTABLE;
 277                 if (inode->i_flags & S_APPEND)
 278                         stat->attributes |= STATX_ATTR_APPEND;
 279         }
 280         return ret;
 281 }
 282
 283 int orangefs_permission(struct inode *inode, int mask)
 284 {
 285         int ret;
 286
 287         if (mask & MAY_NOT_BLOCK)
 288                 return -ECHILD;
 289
 290         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 291
 292         /* Make sure the permission (and other common attrs) are up to date. */
 293         ret = orangefs_inode_getattr(inode, 0, 0, STATX_MODE);
 294         if (ret < 0)
 295                 return ret;
 296
 297         return generic_permission(inode, mask);
 298 }
 299
 300 int orangefs_update_time(struct inode *inode, struct timespec *time, int flags)
 301 {
 302         struct iattr iattr;
 303         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
 304             get_khandle_from_ino(inode));
 305         generic_update_time(inode, time, flags);
 306         memset(&iattr, 0, sizeof iattr);
 307         if (flags & S_ATIME)
 308                 iattr.ia_valid |= ATTR_ATIME;
 309         if (flags & S_CTIME)
 310                 iattr.ia_valid |= ATTR_CTIME;
 311         if (flags & S_MTIME)
 312                 iattr.ia_valid |= ATTR_MTIME;
 313         return orangefs_inode_setattr(inode, &iattr);
 314 }
 315
 316 /* ORANGEDS2 implementation of VFS inode operations for files */
 317 const struct inode_operations orangefs_file_inode_operations = {
 318         .get_acl = orangefs_get_acl,
 319         .set_acl = orangefs_set_acl,
 320         .setattr = orangefs_setattr,
 321         .getattr = orangefs_getattr,
 322         .listxattr = orangefs_listxattr,
 323         .permission = orangefs_permission,
 324         .update_time = orangefs_update_time,
 325 };
 326
 327 static int orangefs_init_iops(struct inode *inode)
 328 {
 329         inode->i_mapping->a_ops = &orangefs_address_operations;
 330
 331         switch (inode->i_mode & S_IFMT) {
 332         case S_IFREG:
 333                 inode->i_op = &orangefs_file_inode_operations;
 334                 inode->i_fop = &orangefs_file_operations;
 335                 inode->i_blkbits = PAGE_SHIFT;
 336                 break;
 337         case S_IFLNK:
 338                 inode->i_op = &orangefs_symlink_inode_operations;
 339                 break;
 340         case S_IFDIR:
 341                 inode->i_op = &orangefs_dir_inode_operations;
 342                 inode->i_fop = &orangefs_dir_operations;
 343                 break;
 344         default:
 345                 gossip_debug(GOSSIP_INODE_DEBUG,
 346                              "%s: unsupported mode\n",
 347                              __func__);
 348                 return -EINVAL;
 349         }
 350
 351         return 0;
 352 }
 353
 354 /*
 355  * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
 356  * that will be used as a hash-index from where the handle will
 357  * be searched for in the VFS hash table of inodes.
 358  */
 359 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
 360 {
 361         if (!ref)
 362                 return 0;
 363         return orangefs_khandle_to_ino(&(ref->khandle));
 364 }
 365
 366 /*
 367  * Called to set up an inode from iget5_locked.
 368  */
 369 static int orangefs_set_inode(struct inode *inode, void *data)
 370 {
 371         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 372         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
 373         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
 374         return 0;
 375 }
 376
 377 /*
 378  * Called to determine if handles match.
 379  */
 380 static int orangefs_test_inode(struct inode *inode, void *data)
 381 {
 382         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 383         struct orangefs_inode_s *orangefs_inode = NULL;
 384
 385         orangefs_inode = ORANGEFS_I(inode);
 386         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle))
 387                 && orangefs_inode->refn.fs_id == ref->fs_id);
 388 }
 389
 390 /*
 391  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
 392  * file handle.
 393  *
 394  * @sb: the file system super block instance.
 395  * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
 396  */
 397 struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref)
 398 {
 399         struct inode *inode = NULL;
 400         unsigned long hash;
 401         int error;
 402
 403         hash = orangefs_handle_hash(ref);
 404         inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref);
 405         if (!inode || !(inode->i_state & I_NEW))
 406                 return inode;
 407
 408         error = orangefs_inode_getattr(inode, 1, 1, STATX_ALL);
 409         if (error) {
 410                 iget_failed(inode);
 411                 return ERR_PTR(error);
 412         }
 413
 414         inode->i_ino = hash;    /* needed for stat etc */
 415         orangefs_init_iops(inode);
 416         unlock_new_inode(inode);
 417
 418         gossip_debug(GOSSIP_INODE_DEBUG,
 419                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
 420                      &ref->khandle,
 421                      ref->fs_id,
 422                      hash,
 423                      inode->i_ino);
 424
 425         return inode;
 426 }
 427
 428 /*
 429  * Allocate an inode for a newly created file and insert it into the inode hash.
 430  */
 431 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
 432                 int mode, dev_t dev, struct orangefs_object_kref *ref)
 433 {
 434         unsigned long hash = orangefs_handle_hash(ref);
 435         struct inode *inode;
 436         int error;
 437
 438         gossip_debug(GOSSIP_INODE_DEBUG,
 439                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
 440                      __func__,
 441                      sb,
 442                      MAJOR(dev),
 443                      MINOR(dev),
 444                      mode);
 445
 446         inode = new_inode(sb);
 447         if (!inode)
 448                 return NULL;
 449
 450         orangefs_set_inode(inode, ref);
 451         inode->i_ino = hash;    /* needed for stat etc */
 452
 453         error = orangefs_inode_getattr(inode, 1, 1, STATX_ALL);
 454         if (error)
 455                 goto out_iput;
 456
 457         orangefs_init_iops(inode);
 458
 459         inode->i_mode = mode;
 460         inode->i_uid = current_fsuid();
 461         inode->i_gid = current_fsgid();
 462         inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 463         inode->i_size = PAGE_SIZE;
 464         inode->i_rdev = dev;
 465
 466         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
 467         if (error < 0)
 468                 goto out_iput;
 469
 470         gossip_debug(GOSSIP_INODE_DEBUG,
 471                      "Initializing ACL's for inode %pU\n",
 472                      get_khandle_from_ino(inode));
 473         orangefs_init_acl(inode, dir);
 474         return inode;
 475
 476 out_iput:
 477         iput(inode);
 478         return ERR_PTR(error);
 479 }