fs/ceph/inode.c

   1 #include "ceph_debug.h"
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/smp_lock.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/namei.h>
  11 #include <linux/writeback.h>
  12 #include <linux/vmalloc.h>
  13 #include <linux/pagevec.h>
  14
  15 #include "super.h"
  16 #include "decode.h"
  17
  18 /*
  19  * Ceph inode operations
  20  *
  21  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  22  * setattr, etc.), xattr helpers, and helpers for assimilating
  23  * metadata returned by the MDS into our cache.
  24  *
  25  * Also define helpers for doing asynchronous writeback, invalidation,
  26  * and truncation for the benefit of those who can't afford to block
  27  * (typically because they are in the message handler path).
  28  */
  29
  30 static const struct inode_operations ceph_symlink_iops;
  31
  32 static void ceph_invalidate_work(struct work_struct *work);
  33 static void ceph_writeback_work(struct work_struct *work);
  34 static void ceph_vmtruncate_work(struct work_struct *work);
  35
  36 /*
  37  * find or create an inode, given the ceph ino number
  38  */
  39 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  40 {
  41         struct inode *inode;
  42         ino_t t = ceph_vino_to_ino(vino);
  43
  44         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  45         if (inode == NULL)
  46                 return ERR_PTR(-ENOMEM);
  47         if (inode->i_state & I_NEW) {
  48                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  49                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  50                 unlock_new_inode(inode);
  51         }
  52
  53         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  54              vino.snap, inode);
  55         return inode;
  56 }
  57
  58 /*
  59  * get/constuct snapdir inode for a given directory
  60  */
  61 struct inode *ceph_get_snapdir(struct inode *parent)
  62 {
  63         struct ceph_vino vino = {
  64                 .ino = ceph_ino(parent),
  65                 .snap = CEPH_SNAPDIR,
  66         };
  67         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  68         struct ceph_inode_info *ci = ceph_inode(inode);
  69
  70         BUG_ON(!S_ISDIR(parent->i_mode));
  71         if (IS_ERR(inode))
  72                 return ERR_PTR(PTR_ERR(inode));
  73         inode->i_mode = parent->i_mode;
  74         inode->i_uid = parent->i_uid;
  75         inode->i_gid = parent->i_gid;
  76         inode->i_op = &ceph_dir_iops;
  77         inode->i_fop = &ceph_dir_fops;
  78         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  79         ci->i_rbytes = 0;
  80         return inode;
  81 }
  82
  83 const struct inode_operations ceph_file_iops = {
  84         .permission = ceph_permission,
  85         .setattr = ceph_setattr,
  86         .getattr = ceph_getattr,
  87         .setxattr = ceph_setxattr,
  88         .getxattr = ceph_getxattr,
  89         .listxattr = ceph_listxattr,
  90         .removexattr = ceph_removexattr,
  91 };
  92
  93
  94 /*
  95  * We use a 'frag tree' to keep track of the MDS's directory fragments
  96  * for a given inode (usually there is just a single fragment).  We
  97  * need to know when a child frag is delegated to a new MDS, or when
  98  * it is flagged as replicated, so we can direct our requests
  99  * accordingly.
 100  */
 101
 102 /*
 103  * find/create a frag in the tree
 104  */
 105 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 106                                                     u32 f)
 107 {
 108         struct rb_node **p;
 109         struct rb_node *parent = NULL;
 110         struct ceph_inode_frag *frag;
 111         int c;
 112
 113         p = &ci->i_fragtree.rb_node;
 114         while (*p) {
 115                 parent = *p;
 116                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 117                 c = ceph_frag_compare(f, frag->frag);
 118                 if (c < 0)
 119                         p = &(*p)->rb_left;
 120                 else if (c > 0)
 121                         p = &(*p)->rb_right;
 122                 else
 123                         return frag;
 124         }
 125
 126         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 127         if (!frag) {
 128                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 129                        "frag %x\n", &ci->vfs_inode,
 130                        ceph_vinop(&ci->vfs_inode), f);
 131                 return ERR_PTR(-ENOMEM);
 132         }
 133         frag->frag = f;
 134         frag->split_by = 0;
 135         frag->mds = -1;
 136         frag->ndist = 0;
 137
 138         rb_link_node(&frag->node, parent, p);
 139         rb_insert_color(&frag->node, &ci->i_fragtree);
 140
 141         dout("get_or_create_frag added %llx.%llx frag %x\n",
 142              ceph_vinop(&ci->vfs_inode), f);
 143         return frag;
 144 }
 145
 146 /*
 147  * find a specific frag @f
 148  */
 149 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 150 {
 151         struct rb_node *n = ci->i_fragtree.rb_node;
 152
 153         while (n) {
 154                 struct ceph_inode_frag *frag =
 155                         rb_entry(n, struct ceph_inode_frag, node);
 156                 int c = ceph_frag_compare(f, frag->frag);
 157                 if (c < 0)
 158                         n = n->rb_left;
 159                 else if (c > 0)
 160                         n = n->rb_right;
 161                 else
 162                         return frag;
 163         }
 164         return NULL;
 165 }
 166
 167 /*
 168  * Choose frag containing the given value @v.  If @pfrag is
 169  * specified, copy the frag delegation info to the caller if
 170  * it is present.
 171  */
 172 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 173                      struct ceph_inode_frag *pfrag,
 174                      int *found)
 175 {
 176         u32 t = ceph_frag_make(0, 0);
 177         struct ceph_inode_frag *frag;
 178         unsigned nway, i;
 179         u32 n;
 180
 181         if (found)
 182                 *found = 0;
 183
 184         mutex_lock(&ci->i_fragtree_mutex);
 185         while (1) {
 186                 WARN_ON(!ceph_frag_contains_value(t, v));
 187                 frag = __ceph_find_frag(ci, t);
 188                 if (!frag)
 189                         break; /* t is a leaf */
 190                 if (frag->split_by == 0) {
 191                         if (pfrag)
 192                                 memcpy(pfrag, frag, sizeof(*pfrag));
 193                         if (found)
 194                                 *found = 1;
 195                         break;
 196                 }
 197
 198                 /* choose child */
 199                 nway = 1 << frag->split_by;
 200                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 201                      frag->split_by, nway);
 202                 for (i = 0; i < nway; i++) {
 203                         n = ceph_frag_make_child(t, frag->split_by, i);
 204                         if (ceph_frag_contains_value(n, v)) {
 205                                 t = n;
 206                                 break;
 207                         }
 208                 }
 209                 BUG_ON(i == nway);
 210         }
 211         dout("choose_frag(%x) = %x\n", v, t);
 212
 213         mutex_unlock(&ci->i_fragtree_mutex);
 214         return t;
 215 }
 216
 217 /*
 218  * Process dirfrag (delegation) info from the mds.  Include leaf
 219  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 220  * branches/splits are included in i_fragtree)
 221  */
 222 static int ceph_fill_dirfrag(struct inode *inode,
 223                              struct ceph_mds_reply_dirfrag *dirinfo)
 224 {
 225         struct ceph_inode_info *ci = ceph_inode(inode);
 226         struct ceph_inode_frag *frag;
 227         u32 id = le32_to_cpu(dirinfo->frag);
 228         int mds = le32_to_cpu(dirinfo->auth);
 229         int ndist = le32_to_cpu(dirinfo->ndist);
 230         int i;
 231         int err = 0;
 232
 233         mutex_lock(&ci->i_fragtree_mutex);
 234         if (ndist == 0) {
 235                 /* no delegation info needed. */
 236                 frag = __ceph_find_frag(ci, id);
 237                 if (!frag)
 238                         goto out;
 239                 if (frag->split_by == 0) {
 240                         /* tree leaf, remove */
 241                         dout("fill_dirfrag removed %llx.%llx frag %x"
 242                              " (no ref)\n", ceph_vinop(inode), id);
 243                         rb_erase(&frag->node, &ci->i_fragtree);
 244                         kfree(frag);
 245                 } else {
 246                         /* tree branch, keep and clear */
 247                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 248                              " referral\n", ceph_vinop(inode), id);
 249                         frag->mds = -1;
 250                         frag->ndist = 0;
 251                 }
 252                 goto out;
 253         }
 254
 255
 256         /* find/add this frag to store mds delegation info */
 257         frag = __get_or_create_frag(ci, id);
 258         if (IS_ERR(frag)) {
 259                 /* this is not the end of the world; we can continue
 260                    with bad/inaccurate delegation info */
 261                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 262                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 263                 err = -ENOMEM;
 264                 goto out;
 265         }
 266
 267         frag->mds = mds;
 268         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 269         for (i = 0; i < frag->ndist; i++)
 270                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 271         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 272              ceph_vinop(inode), frag->frag, frag->ndist);
 273
 274 out:
 275         mutex_unlock(&ci->i_fragtree_mutex);
 276         return err;
 277 }
 278
 279
 280 /*
 281  * initialize a newly allocated inode.
 282  */
 283 struct inode *ceph_alloc_inode(struct super_block *sb)
 284 {
 285         struct ceph_inode_info *ci;
 286         int i;
 287
 288         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 289         if (!ci)
 290                 return NULL;
 291
 292         dout("alloc_inode %p\n", &ci->vfs_inode);
 293
 294         ci->i_version = 0;
 295         ci->i_time_warp_seq = 0;
 296         ci->i_ceph_flags = 0;
 297         ci->i_release_count = 0;
 298         ci->i_symlink = NULL;
 299
 300         ci->i_fragtree = RB_ROOT;
 301         mutex_init(&ci->i_fragtree_mutex);
 302
 303         ci->i_xattrs.blob = NULL;
 304         ci->i_xattrs.prealloc_blob = NULL;
 305         ci->i_xattrs.dirty = false;
 306         ci->i_xattrs.index = RB_ROOT;
 307         ci->i_xattrs.count = 0;
 308         ci->i_xattrs.names_size = 0;
 309         ci->i_xattrs.vals_size = 0;
 310         ci->i_xattrs.version = 0;
 311         ci->i_xattrs.index_version = 0;
 312
 313         ci->i_caps = RB_ROOT;
 314         ci->i_auth_cap = NULL;
 315         ci->i_dirty_caps = 0;
 316         ci->i_flushing_caps = 0;
 317         INIT_LIST_HEAD(&ci->i_dirty_item);
 318         INIT_LIST_HEAD(&ci->i_flushing_item);
 319         ci->i_cap_flush_seq = 0;
 320         ci->i_cap_flush_last_tid = 0;
 321         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 322         init_waitqueue_head(&ci->i_cap_wq);
 323         ci->i_hold_caps_min = 0;
 324         ci->i_hold_caps_max = 0;
 325         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 326         ci->i_cap_exporting_mds = 0;
 327         ci->i_cap_exporting_mseq = 0;
 328         ci->i_cap_exporting_issued = 0;
 329         INIT_LIST_HEAD(&ci->i_cap_snaps);
 330         ci->i_head_snapc = NULL;
 331         ci->i_snap_caps = 0;
 332
 333         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 334                 ci->i_nr_by_mode[i] = 0;
 335
 336         ci->i_truncate_seq = 0;
 337         ci->i_truncate_size = 0;
 338         ci->i_truncate_pending = 0;
 339
 340         ci->i_max_size = 0;
 341         ci->i_reported_size = 0;
 342         ci->i_wanted_max_size = 0;
 343         ci->i_requested_max_size = 0;
 344
 345         ci->i_pin_ref = 0;
 346         ci->i_rd_ref = 0;
 347         ci->i_rdcache_ref = 0;
 348         ci->i_wr_ref = 0;
 349         ci->i_wrbuffer_ref = 0;
 350         ci->i_wrbuffer_ref_head = 0;
 351         ci->i_shared_gen = 0;
 352         ci->i_rdcache_gen = 0;
 353         ci->i_rdcache_revoking = 0;
 354
 355         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 356         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 357         spin_lock_init(&ci->i_unsafe_lock);
 358
 359         ci->i_snap_realm = NULL;
 360         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 361         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 362
 363         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 364         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 365
 366         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 367
 368         return &ci->vfs_inode;
 369 }
 370
 371 void ceph_destroy_inode(struct inode *inode)
 372 {
 373         struct ceph_inode_info *ci = ceph_inode(inode);
 374         struct ceph_inode_frag *frag;
 375         struct rb_node *n;
 376
 377         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 378
 379         ceph_queue_caps_release(inode);
 380
 381         /*
 382          * we may still have a snap_realm reference if there are stray
 383          * caps in i_cap_exporting_issued or i_snap_caps.
 384          */
 385         if (ci->i_snap_realm) {
 386                 struct ceph_mds_client *mdsc =
 387                         &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 388                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 389
 390                 dout(" dropping residual ref to snap realm %p\n", realm);
 391                 spin_lock(&realm->inodes_with_caps_lock);
 392                 list_del_init(&ci->i_snap_realm_item);
 393                 spin_unlock(&realm->inodes_with_caps_lock);
 394                 ceph_put_snap_realm(mdsc, realm);
 395         }
 396
 397         kfree(ci->i_symlink);
 398         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 399                 frag = rb_entry(n, struct ceph_inode_frag, node);
 400                 rb_erase(n, &ci->i_fragtree);
 401                 kfree(frag);
 402         }
 403
 404         __ceph_destroy_xattrs(ci);
 405         if (ci->i_xattrs.blob)
 406                 ceph_buffer_put(ci->i_xattrs.blob);
 407         if (ci->i_xattrs.prealloc_blob)
 408                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 409
 410         kmem_cache_free(ceph_inode_cachep, ci);
 411 }
 412
 413
 414 /*
 415  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 416  * careful because either the client or MDS may have more up to date
 417  * info, depending on which capabilities are held, and whether
 418  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 419  * and size are monotonically increasing, except when utimes() or
 420  * truncate() increments the corresponding _seq values.)
 421  */
 422 int ceph_fill_file_size(struct inode *inode, int issued,
 423                         u32 truncate_seq, u64 truncate_size, u64 size)
 424 {
 425         struct ceph_inode_info *ci = ceph_inode(inode);
 426         int queue_trunc = 0;
 427
 428         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 429             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 430                 dout("size %lld -> %llu\n", inode->i_size, size);
 431                 inode->i_size = size;
 432                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 433                 ci->i_reported_size = size;
 434                 if (truncate_seq != ci->i_truncate_seq) {
 435                         dout("truncate_seq %u -> %u\n",
 436                              ci->i_truncate_seq, truncate_seq);
 437                         ci->i_truncate_seq = truncate_seq;
 438                         /*
 439                          * If we hold relevant caps, or in the case where we're
 440                          * not the only client referencing this file and we
 441                          * don't hold those caps, then we need to check whether
 442                          * the file is either opened or mmaped
 443                          */
 444                         if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
 445                                       CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
 446                                       CEPH_CAP_FILE_EXCL)) ||
 447                             mapping_mapped(inode->i_mapping) ||
 448                             __ceph_caps_file_wanted(ci)) {
 449                                 ci->i_truncate_pending++;
 450                                 queue_trunc = 1;
 451                         }
 452                 }
 453         }
 454         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 455             ci->i_truncate_size != truncate_size) {
 456                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 457                      truncate_size);
 458                 ci->i_truncate_size = truncate_size;
 459         }
 460         return queue_trunc;
 461 }
 462
 463 void ceph_fill_file_time(struct inode *inode, int issued,
 464                          u64 time_warp_seq, struct timespec *ctime,
 465                          struct timespec *mtime, struct timespec *atime)
 466 {
 467         struct ceph_inode_info *ci = ceph_inode(inode);
 468         int warn = 0;
 469
 470         if (issued & (CEPH_CAP_FILE_EXCL|
 471                       CEPH_CAP_FILE_WR|
 472                       CEPH_CAP_FILE_BUFFER)) {
 473                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 474                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 475                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 476                              ctime->tv_sec, ctime->tv_nsec);
 477                         inode->i_ctime = *ctime;
 478                 }
 479                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 480                         /* the MDS did a utimes() */
 481                         dout("mtime %ld.%09ld -> %ld.%09ld "
 482                              "tw %d -> %d\n",
 483                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 484                              mtime->tv_sec, mtime->tv_nsec,
 485                              ci->i_time_warp_seq, (int)time_warp_seq);
 486
 487                         inode->i_mtime = *mtime;
 488                         inode->i_atime = *atime;
 489                         ci->i_time_warp_seq = time_warp_seq;
 490                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 491                         /* nobody did utimes(); take the max */
 492                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 493                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 494                                      inode->i_mtime.tv_sec,
 495                                      inode->i_mtime.tv_nsec,
 496                                      mtime->tv_sec, mtime->tv_nsec);
 497                                 inode->i_mtime = *mtime;
 498                         }
 499                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 500                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 501                                      inode->i_atime.tv_sec,
 502                                      inode->i_atime.tv_nsec,
 503                                      atime->tv_sec, atime->tv_nsec);
 504                                 inode->i_atime = *atime;
 505                         }
 506                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 507                         /* we did a utimes(); ignore mds values */
 508                 } else {
 509                         warn = 1;
 510                 }
 511         } else {
 512                 /* we have no write caps; whatever the MDS says is true */
 513                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 514                         inode->i_ctime = *ctime;
 515                         inode->i_mtime = *mtime;
 516                         inode->i_atime = *atime;
 517                         ci->i_time_warp_seq = time_warp_seq;
 518                 } else {
 519                         warn = 1;
 520                 }
 521         }
 522         if (warn) /* time_warp_seq shouldn't go backwards */
 523                 dout("%p mds time_warp_seq %llu < %u\n",
 524                      inode, time_warp_seq, ci->i_time_warp_seq);
 525 }
 526
 527 /*
 528  * Populate an inode based on info from mds.  May be called on new or
 529  * existing inodes.
 530  */
 531 static int fill_inode(struct inode *inode,
 532                       struct ceph_mds_reply_info_in *iinfo,
 533                       struct ceph_mds_reply_dirfrag *dirinfo,
 534                       struct ceph_mds_session *session,
 535                       unsigned long ttl_from, int cap_fmode,
 536                       struct ceph_cap_reservation *caps_reservation)
 537 {
 538         struct ceph_mds_reply_inode *info = iinfo->in;
 539         struct ceph_inode_info *ci = ceph_inode(inode);
 540         int i;
 541         int issued, implemented;
 542         struct timespec mtime, atime, ctime;
 543         u32 nsplits;
 544         struct ceph_buffer *xattr_blob = NULL;
 545         int err = 0;
 546         int queue_trunc = 0;
 547
 548         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 549              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 550              ci->i_version);
 551
 552         /*
 553          * prealloc xattr data, if it looks like we'll need it.  only
 554          * if len > 4 (meaning there are actually xattrs; the first 4
 555          * bytes are the xattr count).
 556          */
 557         if (iinfo->xattr_len > 4) {
 558                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 559                 if (!xattr_blob)
 560                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 561                                iinfo->xattr_len);
 562         }
 563
 564         spin_lock(&inode->i_lock);
 565
 566         /*
 567          * provided version will be odd if inode value is projected,
 568          * even if stable.  skip the update if we have a newer info
 569          * (e.g., due to inode info racing form multiple MDSs), or if
 570          * we are getting projected (unstable) inode info.
 571          */
 572         if (le64_to_cpu(info->version) > 0 &&
 573             (ci->i_version & ~1) > le64_to_cpu(info->version))
 574                 goto no_change;
 575
 576         issued = __ceph_caps_issued(ci, &implemented);
 577         issued |= implemented | __ceph_caps_dirty(ci);
 578
 579         /* update inode */
 580         ci->i_version = le64_to_cpu(info->version);
 581         inode->i_version++;
 582         inode->i_rdev = le32_to_cpu(info->rdev);
 583
 584         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
 585                 inode->i_mode = le32_to_cpu(info->mode);
 586                 inode->i_uid = le32_to_cpu(info->uid);
 587                 inode->i_gid = le32_to_cpu(info->gid);
 588                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 589                      inode->i_uid, inode->i_gid);
 590         }
 591
 592         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
 593                 inode->i_nlink = le32_to_cpu(info->nlink);
 594
 595         /* be careful with mtime, atime, size */
 596         ceph_decode_timespec(&atime, &info->atime);
 597         ceph_decode_timespec(&mtime, &info->mtime);
 598         ceph_decode_timespec(&ctime, &info->ctime);
 599         queue_trunc = ceph_fill_file_size(inode, issued,
 600                                           le32_to_cpu(info->truncate_seq),
 601                                           le64_to_cpu(info->truncate_size),
 602                                           le64_to_cpu(info->size));
 603         ceph_fill_file_time(inode, issued,
 604                             le32_to_cpu(info->time_warp_seq),
 605                             &ctime, &mtime, &atime);
 606
 607         ci->i_max_size = le64_to_cpu(info->max_size);
 608         ci->i_layout = info->layout;
 609         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 610
 611         /* xattrs */
 612         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 613         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
 614             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 615                 if (ci->i_xattrs.blob)
 616                         ceph_buffer_put(ci->i_xattrs.blob);
 617                 ci->i_xattrs.blob = xattr_blob;
 618                 if (xattr_blob)
 619                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 620                                iinfo->xattr_data, iinfo->xattr_len);
 621                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 622                 xattr_blob = NULL;
 623         }
 624
 625         inode->i_mapping->a_ops = &ceph_aops;
 626         inode->i_mapping->backing_dev_info =
 627                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
 628
 629         switch (inode->i_mode & S_IFMT) {
 630         case S_IFIFO:
 631         case S_IFBLK:
 632         case S_IFCHR:
 633         case S_IFSOCK:
 634                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 635                 inode->i_op = &ceph_file_iops;
 636                 break;
 637         case S_IFREG:
 638                 inode->i_op = &ceph_file_iops;
 639                 inode->i_fop = &ceph_file_fops;
 640                 break;
 641         case S_IFLNK:
 642                 inode->i_op = &ceph_symlink_iops;
 643                 if (!ci->i_symlink) {
 644                         int symlen = iinfo->symlink_len;
 645                         char *sym;
 646
 647                         BUG_ON(symlen != inode->i_size);
 648                         spin_unlock(&inode->i_lock);
 649
 650                         err = -ENOMEM;
 651                         sym = kmalloc(symlen+1, GFP_NOFS);
 652                         if (!sym)
 653                                 goto out;
 654                         memcpy(sym, iinfo->symlink, symlen);
 655                         sym[symlen] = 0;
 656
 657                         spin_lock(&inode->i_lock);
 658                         if (!ci->i_symlink)
 659                                 ci->i_symlink = sym;
 660                         else
 661                                 kfree(sym); /* lost a race */
 662                 }
 663                 break;
 664         case S_IFDIR:
 665                 inode->i_op = &ceph_dir_iops;
 666                 inode->i_fop = &ceph_dir_fops;
 667
 668                 ci->i_files = le64_to_cpu(info->files);
 669                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 670                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 671                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 672                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 673                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 674
 675                 /* set dir completion flag? */
 676                 if (ci->i_files == 0 && ci->i_subdirs == 0 &&
 677                     ceph_snap(inode) == CEPH_NOSNAP &&
 678                     (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
 679                     (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
 680                         dout(" marking %p complete (empty)\n", inode);
 681                         ci->i_ceph_flags |= CEPH_I_COMPLETE;
 682                         ci->i_max_offset = 2;
 683                 }
 684
 685                 /* it may be better to set st_size in getattr instead? */
 686                 if (ceph_test_opt(ceph_sb_to_client(inode->i_sb), RBYTES))
 687                         inode->i_size = ci->i_rbytes;
 688                 break;
 689         default:
 690                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 691                        ceph_vinop(inode), inode->i_mode);
 692         }
 693
 694 no_change:
 695         spin_unlock(&inode->i_lock);
 696
 697         /* queue truncate if we saw i_size decrease */
 698         if (queue_trunc)
 699                 ceph_queue_vmtruncate(inode);
 700
 701         /* populate frag tree */
 702         /* FIXME: move me up, if/when version reflects fragtree changes */
 703         nsplits = le32_to_cpu(info->fragtree.nsplits);
 704         mutex_lock(&ci->i_fragtree_mutex);
 705         for (i = 0; i < nsplits; i++) {
 706                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
 707                 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
 708
 709                 if (IS_ERR(frag))
 710                         continue;
 711                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
 712                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 713         }
 714         mutex_unlock(&ci->i_fragtree_mutex);
 715
 716         /* were we issued a capability? */
 717         if (info->cap.caps) {
 718                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 719                         ceph_add_cap(inode, session,
 720                                      le64_to_cpu(info->cap.cap_id),
 721                                      cap_fmode,
 722                                      le32_to_cpu(info->cap.caps),
 723                                      le32_to_cpu(info->cap.wanted),
 724                                      le32_to_cpu(info->cap.seq),
 725                                      le32_to_cpu(info->cap.mseq),
 726                                      le64_to_cpu(info->cap.realm),
 727                                      info->cap.flags,
 728                                      caps_reservation);
 729                 } else {
 730                         spin_lock(&inode->i_lock);
 731                         dout(" %p got snap_caps %s\n", inode,
 732                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 733                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 734                         if (cap_fmode >= 0)
 735                                 __ceph_get_fmode(ci, cap_fmode);
 736                         spin_unlock(&inode->i_lock);
 737                 }
 738         } else if (cap_fmode >= 0) {
 739                 pr_warning("mds issued no caps on %llx.%llx\n",
 740                            ceph_vinop(inode));
 741                 __ceph_get_fmode(ci, cap_fmode);
 742         }
 743
 744         /* update delegation info? */
 745         if (dirinfo)
 746                 ceph_fill_dirfrag(inode, dirinfo);
 747
 748         err = 0;
 749
 750 out:
 751         if (xattr_blob)
 752                 ceph_buffer_put(xattr_blob);
 753         return err;
 754 }
 755
 756 /*
 757  * caller should hold session s_mutex.
 758  */
 759 static void update_dentry_lease(struct dentry *dentry,
 760                                 struct ceph_mds_reply_lease *lease,
 761                                 struct ceph_mds_session *session,
 762                                 unsigned long from_time)
 763 {
 764         struct ceph_dentry_info *di = ceph_dentry(dentry);
 765         long unsigned duration = le32_to_cpu(lease->duration_ms);
 766         long unsigned ttl = from_time + (duration * HZ) / 1000;
 767         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 768         struct inode *dir;
 769
 770         /* only track leases on regular dentries */
 771         if (dentry->d_op != &ceph_dentry_ops)
 772                 return;
 773
 774         spin_lock(&dentry->d_lock);
 775         dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
 776              dentry, le16_to_cpu(lease->mask), duration, ttl);
 777
 778         /* make lease_rdcache_gen match directory */
 779         dir = dentry->d_parent->d_inode;
 780         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 781
 782         if (lease->mask == 0)
 783                 goto out_unlock;
 784
 785         if (di->lease_gen == session->s_cap_gen &&
 786             time_before(ttl, dentry->d_time))
 787                 goto out_unlock;  /* we already have a newer lease. */
 788
 789         if (di->lease_session && di->lease_session != session)
 790                 goto out_unlock;
 791
 792         ceph_dentry_lru_touch(dentry);
 793
 794         if (!di->lease_session)
 795                 di->lease_session = ceph_get_mds_session(session);
 796         di->lease_gen = session->s_cap_gen;
 797         di->lease_seq = le32_to_cpu(lease->seq);
 798         di->lease_renew_after = half_ttl;
 799         di->lease_renew_from = 0;
 800         dentry->d_time = ttl;
 801 out_unlock:
 802         spin_unlock(&dentry->d_lock);
 803         return;
 804 }
 805
 806 /*
 807  * Set dentry's directory position based on the current dir's max, and
 808  * order it in d_subdirs, so that dcache_readdir behaves.
 809  */
 810 static void ceph_set_dentry_offset(struct dentry *dn)
 811 {
 812         struct dentry *dir = dn->d_parent;
 813         struct inode *inode = dn->d_parent->d_inode;
 814         struct ceph_dentry_info *di;
 815
 816         BUG_ON(!inode);
 817
 818         di = ceph_dentry(dn);
 819
 820         spin_lock(&inode->i_lock);
 821         if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
 822                 spin_unlock(&inode->i_lock);
 823                 return;
 824         }
 825         di->offset = ceph_inode(inode)->i_max_offset++;
 826         spin_unlock(&inode->i_lock);
 827
 828         spin_lock(&dcache_lock);
 829         spin_lock(&dn->d_lock);
 830         list_move_tail(&dir->d_subdirs, &dn->d_u.d_child);
 831         dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
 832              dn->d_u.d_child.prev, dn->d_u.d_child.next);
 833         spin_unlock(&dn->d_lock);
 834         spin_unlock(&dcache_lock);
 835 }
 836
 837 /*
 838  * splice a dentry to an inode.
 839  * caller must hold directory i_mutex for this to be safe.
 840  *
 841  * we will only rehash the resulting dentry if @prehash is
 842  * true; @prehash will be set to false (for the benefit of
 843  * the caller) if we fail.
 844  */
 845 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 846                                     bool *prehash)
 847 {
 848         struct dentry *realdn;
 849
 850         BUG_ON(dn->d_inode);
 851
 852         /* dn must be unhashed */
 853         if (!d_unhashed(dn))
 854                 d_drop(dn);
 855         realdn = d_materialise_unique(dn, in);
 856         if (IS_ERR(realdn)) {
 857                 pr_err("splice_dentry error %p inode %p ino %llx.%llx\n",
 858                        dn, in, ceph_vinop(in));
 859                 if (prehash)
 860                         *prehash = false; /* don't rehash on error */
 861                 dn = realdn; /* note realdn contains the error */
 862                 goto out;
 863         } else if (realdn) {
 864                 dout("dn %p (%d) spliced with %p (%d) "
 865                      "inode %p ino %llx.%llx\n",
 866                      dn, atomic_read(&dn->d_count),
 867                      realdn, atomic_read(&realdn->d_count),
 868                      realdn->d_inode, ceph_vinop(realdn->d_inode));
 869                 dput(dn);
 870                 dn = realdn;
 871         } else {
 872                 BUG_ON(!ceph_dentry(dn));
 873                 dout("dn %p attached to %p ino %llx.%llx\n",
 874                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
 875         }
 876         if ((!prehash || *prehash) && d_unhashed(dn))
 877                 d_rehash(dn);
 878         ceph_set_dentry_offset(dn);
 879 out:
 880         return dn;
 881 }
 882
 883 /*
 884  * Incorporate results into the local cache.  This is either just
 885  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
 886  * after a lookup).
 887  *
 888  * A reply may contain
 889  *         a directory inode along with a dentry.
 890  *  and/or a target inode
 891  *
 892  * Called with snap_rwsem (read).
 893  */
 894 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
 895                     struct ceph_mds_session *session)
 896 {
 897         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
 898         struct inode *in = NULL;
 899         struct ceph_mds_reply_inode *ininfo;
 900         struct ceph_vino vino;
 901         struct ceph_client *client = ceph_sb_to_client(sb);
 902         int i = 0;
 903         int err = 0;
 904
 905         dout("fill_trace %p is_dentry %d is_target %d\n", req,
 906              rinfo->head->is_dentry, rinfo->head->is_target);
 907
 908 #if 0
 909         /*
 910          * Debugging hook:
 911          *
 912          * If we resend completed ops to a recovering mds, we get no
 913          * trace.  Since that is very rare, pretend this is the case
 914          * to ensure the 'no trace' handlers in the callers behave.
 915          *
 916          * Fill in inodes unconditionally to avoid breaking cap
 917          * invariants.
 918          */
 919         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
 920                 pr_info("fill_trace faking empty trace on %lld %s\n",
 921                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
 922                 if (rinfo->head->is_dentry) {
 923                         rinfo->head->is_dentry = 0;
 924                         err = fill_inode(req->r_locked_dir,
 925                                          &rinfo->diri, rinfo->dirfrag,
 926                                          session, req->r_request_started, -1);
 927                 }
 928                 if (rinfo->head->is_target) {
 929                         rinfo->head->is_target = 0;
 930                         ininfo = rinfo->targeti.in;
 931                         vino.ino = le64_to_cpu(ininfo->ino);
 932                         vino.snap = le64_to_cpu(ininfo->snapid);
 933                         in = ceph_get_inode(sb, vino);
 934                         err = fill_inode(in, &rinfo->targeti, NULL,
 935                                          session, req->r_request_started,
 936                                          req->r_fmode);
 937                         iput(in);
 938                 }
 939         }
 940 #endif
 941
 942         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
 943                 dout("fill_trace reply is empty!\n");
 944                 if (rinfo->head->result == 0 && req->r_locked_dir) {
 945                         struct ceph_inode_info *ci =
 946                                 ceph_inode(req->r_locked_dir);
 947                         dout(" clearing %p complete (empty trace)\n",
 948                              req->r_locked_dir);
 949                         spin_lock(&req->r_locked_dir->i_lock);
 950                         ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
 951                         ci->i_release_count++;
 952                         spin_unlock(&req->r_locked_dir->i_lock);
 953
 954                         if (req->r_dentry)
 955                                 ceph_invalidate_dentry_lease(req->r_dentry);
 956                         if (req->r_old_dentry)
 957                                 ceph_invalidate_dentry_lease(req->r_old_dentry);
 958                 }
 959                 return 0;
 960         }
 961
 962         if (rinfo->head->is_dentry) {
 963                 struct inode *dir = req->r_locked_dir;
 964
 965                 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
 966                                  session, req->r_request_started, -1,
 967                                  &req->r_caps_reservation);
 968                 if (err < 0)
 969                         return err;
 970         }
 971
 972         /*
 973          * ignore null lease/binding on snapdir ENOENT, or else we
 974          * will have trouble splicing in the virtual snapdir later
 975          */
 976         if (rinfo->head->is_dentry && !req->r_aborted &&
 977             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
 978                                                client->mount_args->snapdir_name,
 979                                                req->r_dentry->d_name.len))) {
 980                 /*
 981                  * lookup link rename   : null -> possibly existing inode
 982                  * mknod symlink mkdir  : null -> new inode
 983                  * unlink               : linked -> null
 984                  */
 985                 struct inode *dir = req->r_locked_dir;
 986                 struct dentry *dn = req->r_dentry;
 987                 bool have_dir_cap, have_lease;
 988
 989                 BUG_ON(!dn);
 990                 BUG_ON(!dir);
 991                 BUG_ON(dn->d_parent->d_inode != dir);
 992                 BUG_ON(ceph_ino(dir) !=
 993                        le64_to_cpu(rinfo->diri.in->ino));
 994                 BUG_ON(ceph_snap(dir) !=
 995                        le64_to_cpu(rinfo->diri.in->snapid));
 996
 997                 /* do we have a lease on the whole dir? */
 998                 have_dir_cap =
 999                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1000                          CEPH_CAP_FILE_SHARED);
1001
1002                 /* do we have a dn lease? */
1003                 have_lease = have_dir_cap ||
1004                         (le16_to_cpu(rinfo->dlease->mask) &
1005                          CEPH_LOCK_DN);
1006
1007                 if (!have_lease)
1008                         dout("fill_trace  no dentry lease or dir cap\n");
1009
1010                 /* rename? */
1011                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1012                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1013                              req->r_old_dentry,
1014                              req->r_old_dentry->d_name.len,
1015                              req->r_old_dentry->d_name.name,
1016                              dn, dn->d_name.len, dn->d_name.name);
1017                         dout("fill_trace doing d_move %p -> %p\n",
1018                              req->r_old_dentry, dn);
1019
1020                         /* d_move screws up d_subdirs order */
1021                         ceph_i_clear(dir, CEPH_I_COMPLETE);
1022
1023                         d_move(req->r_old_dentry, dn);
1024                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1025                              req->r_old_dentry,
1026                              req->r_old_dentry->d_name.len,
1027                              req->r_old_dentry->d_name.name,
1028                              dn, dn->d_name.len, dn->d_name.name);
1029
1030                         /* ensure target dentry is invalidated, despite
1031                            rehashing bug in vfs_rename_dir */
1032                         ceph_invalidate_dentry_lease(dn);
1033
1034                         /* take overwritten dentry's readdir offset */
1035                         dout("dn %p gets %p offset %lld (old offset %lld)\n",
1036                              req->r_old_dentry, dn, ceph_dentry(dn)->offset,
1037                              ceph_dentry(req->r_old_dentry)->offset);
1038                         ceph_dentry(req->r_old_dentry)->offset =
1039                                 ceph_dentry(dn)->offset;
1040
1041                         dn = req->r_old_dentry;  /* use old_dentry */
1042                         in = dn->d_inode;
1043                 }
1044
1045                 /* null dentry? */
1046                 if (!rinfo->head->is_target) {
1047                         dout("fill_trace null dentry\n");
1048                         if (dn->d_inode) {
1049                                 dout("d_delete %p\n", dn);
1050                                 d_delete(dn);
1051                         } else {
1052                                 dout("d_instantiate %p NULL\n", dn);
1053                                 d_instantiate(dn, NULL);
1054                                 if (have_lease && d_unhashed(dn))
1055                                         d_rehash(dn);
1056                                 update_dentry_lease(dn, rinfo->dlease,
1057                                                     session,
1058                                                     req->r_request_started);
1059                         }
1060                         goto done;
1061                 }
1062
1063                 /* attach proper inode */
1064                 ininfo = rinfo->targeti.in;
1065                 vino.ino = le64_to_cpu(ininfo->ino);
1066                 vino.snap = le64_to_cpu(ininfo->snapid);
1067                 if (!dn->d_inode) {
1068                         in = ceph_get_inode(sb, vino);
1069                         if (IS_ERR(in)) {
1070                                 pr_err("fill_trace bad get_inode "
1071                                        "%llx.%llx\n", vino.ino, vino.snap);
1072                                 err = PTR_ERR(in);
1073                                 d_delete(dn);
1074                                 goto done;
1075                         }
1076                         dn = splice_dentry(dn, in, &have_lease);
1077                         if (IS_ERR(dn)) {
1078                                 err = PTR_ERR(dn);
1079                                 goto done;
1080                         }
1081                         req->r_dentry = dn;  /* may have spliced */
1082                         igrab(in);
1083                 } else if (ceph_ino(in) == vino.ino &&
1084                            ceph_snap(in) == vino.snap) {
1085                         igrab(in);
1086                 } else {
1087                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1088                              dn, in, ceph_ino(in), ceph_snap(in),
1089                              vino.ino, vino.snap);
1090                         have_lease = false;
1091                         in = NULL;
1092                 }
1093
1094                 if (have_lease)
1095                         update_dentry_lease(dn, rinfo->dlease, session,
1096                                             req->r_request_started);
1097                 dout(" final dn %p\n", dn);
1098                 i++;
1099         } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1100                    req->r_op == CEPH_MDS_OP_MKSNAP) {
1101                 struct dentry *dn = req->r_dentry;
1102
1103                 /* fill out a snapdir LOOKUPSNAP dentry */
1104                 BUG_ON(!dn);
1105                 BUG_ON(!req->r_locked_dir);
1106                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1107                 ininfo = rinfo->targeti.in;
1108                 vino.ino = le64_to_cpu(ininfo->ino);
1109                 vino.snap = le64_to_cpu(ininfo->snapid);
1110                 in = ceph_get_inode(sb, vino);
1111                 if (IS_ERR(in)) {
1112                         pr_err("fill_inode get_inode badness %llx.%llx\n",
1113                                vino.ino, vino.snap);
1114                         err = PTR_ERR(in);
1115                         d_delete(dn);
1116                         goto done;
1117                 }
1118                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1119                 dn = splice_dentry(dn, in, NULL);
1120                 if (IS_ERR(dn)) {
1121                         err = PTR_ERR(dn);
1122                         goto done;
1123                 }
1124                 req->r_dentry = dn;  /* may have spliced */
1125                 igrab(in);
1126                 rinfo->head->is_dentry = 1;  /* fool notrace handlers */
1127         }
1128
1129         if (rinfo->head->is_target) {
1130                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1131                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1132
1133                 if (in == NULL || ceph_ino(in) != vino.ino ||
1134                     ceph_snap(in) != vino.snap) {
1135                         in = ceph_get_inode(sb, vino);
1136                         if (IS_ERR(in)) {
1137                                 err = PTR_ERR(in);
1138                                 goto done;
1139                         }
1140                 }
1141                 req->r_target_inode = in;
1142
1143                 err = fill_inode(in,
1144                                  &rinfo->targeti, NULL,
1145                                  session, req->r_request_started,
1146                                  (le32_to_cpu(rinfo->head->result) == 0) ?
1147                                  req->r_fmode : -1,
1148                                  &req->r_caps_reservation);
1149                 if (err < 0) {
1150                         pr_err("fill_inode badness %p %llx.%llx\n",
1151                                in, ceph_vinop(in));
1152                         goto done;
1153                 }
1154         }
1155
1156 done:
1157         dout("fill_trace done err=%d\n", err);
1158         return err;
1159 }
1160
1161 /*
1162  * Prepopulate our cache with readdir results, leases, etc.
1163  */
1164 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1165                              struct ceph_mds_session *session)
1166 {
1167         struct dentry *parent = req->r_dentry;
1168         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1169         struct qstr dname;
1170         struct dentry *dn;
1171         struct inode *in;
1172         int err = 0, i;
1173         struct inode *snapdir = NULL;
1174         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1175         u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1176         struct ceph_dentry_info *di;
1177
1178         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1179                 snapdir = ceph_get_snapdir(parent->d_inode);
1180                 parent = d_find_alias(snapdir);
1181                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1182                      rinfo->dir_nr, parent);
1183         } else {
1184                 dout("readdir_prepopulate %d items under dn %p\n",
1185                      rinfo->dir_nr, parent);
1186                 if (rinfo->dir_dir)
1187                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1188         }
1189
1190         for (i = 0; i < rinfo->dir_nr; i++) {
1191                 struct ceph_vino vino;
1192
1193                 dname.name = rinfo->dir_dname[i];
1194                 dname.len = rinfo->dir_dname_len[i];
1195                 dname.hash = full_name_hash(dname.name, dname.len);
1196
1197                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1198                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1199
1200 retry_lookup:
1201                 dn = d_lookup(parent, &dname);
1202                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1203                      parent, dname.len, dname.name, dn);
1204
1205                 if (!dn) {
1206                         dn = d_alloc(parent, &dname);
1207                         dout("d_alloc %p '%.*s' = %p\n", parent,
1208                              dname.len, dname.name, dn);
1209                         if (dn == NULL) {
1210                                 dout("d_alloc badness\n");
1211                                 err = -ENOMEM;
1212                                 goto out;
1213                         }
1214                         err = ceph_init_dentry(dn);
1215                         if (err < 0)
1216                                 goto out;
1217                 } else if (dn->d_inode &&
1218                            (ceph_ino(dn->d_inode) != vino.ino ||
1219                             ceph_snap(dn->d_inode) != vino.snap)) {
1220                         dout(" dn %p points to wrong inode %p\n",
1221                              dn, dn->d_inode);
1222                         d_delete(dn);
1223                         dput(dn);
1224                         goto retry_lookup;
1225                 } else {
1226                         /* reorder parent's d_subdirs */
1227                         spin_lock(&dcache_lock);
1228                         spin_lock(&dn->d_lock);
1229                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1230                         spin_unlock(&dn->d_lock);
1231                         spin_unlock(&dcache_lock);
1232                 }
1233
1234                 di = dn->d_fsdata;
1235                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1236
1237                 /* inode */
1238                 if (dn->d_inode) {
1239                         in = dn->d_inode;
1240                 } else {
1241                         in = ceph_get_inode(parent->d_sb, vino);
1242                         if (in == NULL) {
1243                                 dout("new_inode badness\n");
1244                                 d_delete(dn);
1245                                 dput(dn);
1246                                 err = -ENOMEM;
1247                                 goto out;
1248                         }
1249                         dn = splice_dentry(dn, in, NULL);
1250                 }
1251
1252                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1253                                req->r_request_started, -1,
1254                                &req->r_caps_reservation) < 0) {
1255                         pr_err("fill_inode badness on %p\n", in);
1256                         dput(dn);
1257                         continue;
1258                 }
1259                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1260                                     req->r_session, req->r_request_started);
1261                 dput(dn);
1262         }
1263         req->r_did_prepopulate = true;
1264
1265 out:
1266         if (snapdir) {
1267                 iput(snapdir);
1268                 dput(parent);
1269         }
1270         dout("readdir_prepopulate done\n");
1271         return err;
1272 }
1273
1274 int ceph_inode_set_size(struct inode *inode, loff_t size)
1275 {
1276         struct ceph_inode_info *ci = ceph_inode(inode);
1277         int ret = 0;
1278
1279         spin_lock(&inode->i_lock);
1280         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1281         inode->i_size = size;
1282         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1283
1284         /* tell the MDS if we are approaching max_size */
1285         if ((size << 1) >= ci->i_max_size &&
1286             (ci->i_reported_size << 1) < ci->i_max_size)
1287                 ret = 1;
1288
1289         spin_unlock(&inode->i_lock);
1290         return ret;
1291 }
1292
1293 /*
1294  * Write back inode data in a worker thread.  (This can't be done
1295  * in the message handler context.)
1296  */
1297 void ceph_queue_writeback(struct inode *inode)
1298 {
1299         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1300                        &ceph_inode(inode)->i_wb_work)) {
1301                 dout("ceph_queue_writeback %p\n", inode);
1302                 igrab(inode);
1303         } else {
1304                 dout("ceph_queue_writeback %p failed\n", inode);
1305         }
1306 }
1307
1308 static void ceph_writeback_work(struct work_struct *work)
1309 {
1310         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1311                                                   i_wb_work);
1312         struct inode *inode = &ci->vfs_inode;
1313
1314         dout("writeback %p\n", inode);
1315         filemap_fdatawrite(&inode->i_data);
1316         iput(inode);
1317 }
1318
1319 /*
1320  * queue an async invalidation
1321  */
1322 void ceph_queue_invalidate(struct inode *inode)
1323 {
1324         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1325                        &ceph_inode(inode)->i_pg_inv_work)) {
1326                 dout("ceph_queue_invalidate %p\n", inode);
1327                 igrab(inode);
1328         } else {
1329                 dout("ceph_queue_invalidate %p failed\n", inode);
1330         }
1331 }
1332
1333 /*
1334  * invalidate any pages that are not dirty or under writeback.  this
1335  * includes pages that are clean and mapped.
1336  */
1337 static void ceph_invalidate_nondirty_pages(struct address_space *mapping)
1338 {
1339         struct pagevec pvec;
1340         pgoff_t next = 0;
1341         int i;
1342
1343         pagevec_init(&pvec, 0);
1344         while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
1345                 for (i = 0; i < pagevec_count(&pvec); i++) {
1346                         struct page *page = pvec.pages[i];
1347                         pgoff_t index;
1348                         int skip_page =
1349                                 (PageDirty(page) || PageWriteback(page));
1350
1351                         if (!skip_page)
1352                                 skip_page = !trylock_page(page);
1353
1354                         /*
1355                          * We really shouldn't be looking at the ->index of an
1356                          * unlocked page.  But we're not allowed to lock these
1357                          * pages.  So we rely upon nobody altering the ->index
1358                          * of this (pinned-by-us) page.
1359                          */
1360                         index = page->index;
1361                         if (index > next)
1362                                 next = index;
1363                         next++;
1364
1365                         if (skip_page)
1366                                 continue;
1367
1368                         generic_error_remove_page(mapping, page);
1369                         unlock_page(page);
1370                 }
1371                 pagevec_release(&pvec);
1372                 cond_resched();
1373         }
1374 }
1375
1376 /*
1377  * Invalidate inode pages in a worker thread.  (This can't be done
1378  * in the message handler context.)
1379  */
1380 static void ceph_invalidate_work(struct work_struct *work)
1381 {
1382         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1383                                                   i_pg_inv_work);
1384         struct inode *inode = &ci->vfs_inode;
1385         u32 orig_gen;
1386         int check = 0;
1387
1388         spin_lock(&inode->i_lock);
1389         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1390              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1391         if (ci->i_rdcache_gen == 0 ||
1392             ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1393                 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen);
1394                 /* nevermind! */
1395                 ci->i_rdcache_revoking = 0;
1396                 spin_unlock(&inode->i_lock);
1397                 goto out;
1398         }
1399         orig_gen = ci->i_rdcache_gen;
1400         spin_unlock(&inode->i_lock);
1401
1402         ceph_invalidate_nondirty_pages(inode->i_mapping);
1403
1404         spin_lock(&inode->i_lock);
1405         if (orig_gen == ci->i_rdcache_gen) {
1406                 dout("invalidate_pages %p gen %d successful\n", inode,
1407                      ci->i_rdcache_gen);
1408                 ci->i_rdcache_gen = 0;
1409                 ci->i_rdcache_revoking = 0;
1410                 check = 1;
1411         } else {
1412                 dout("invalidate_pages %p gen %d raced, gen now %d\n",
1413                      inode, orig_gen, ci->i_rdcache_gen);
1414         }
1415         spin_unlock(&inode->i_lock);
1416
1417         if (check)
1418                 ceph_check_caps(ci, 0, NULL);
1419 out:
1420         iput(inode);
1421 }
1422
1423
1424 /*
1425  * called by trunc_wq; take i_mutex ourselves
1426  *
1427  * We also truncate in a separate thread as well.
1428  */
1429 static void ceph_vmtruncate_work(struct work_struct *work)
1430 {
1431         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1432                                                   i_vmtruncate_work);
1433         struct inode *inode = &ci->vfs_inode;
1434
1435         dout("vmtruncate_work %p\n", inode);
1436         mutex_lock(&inode->i_mutex);
1437         __ceph_do_pending_vmtruncate(inode);
1438         mutex_unlock(&inode->i_mutex);
1439         iput(inode);
1440 }
1441
1442 /*
1443  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1444  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1445  */
1446 void ceph_queue_vmtruncate(struct inode *inode)
1447 {
1448         struct ceph_inode_info *ci = ceph_inode(inode);
1449
1450         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1451                        &ci->i_vmtruncate_work)) {
1452                 dout("ceph_queue_vmtruncate %p\n", inode);
1453                 igrab(inode);
1454         } else {
1455                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1456                      inode, ci->i_truncate_pending);
1457         }
1458 }
1459
1460 /*
1461  * called with i_mutex held.
1462  *
1463  * Make sure any pending truncation is applied before doing anything
1464  * that may depend on it.
1465  */
1466 void __ceph_do_pending_vmtruncate(struct inode *inode)
1467 {
1468         struct ceph_inode_info *ci = ceph_inode(inode);
1469         u64 to;
1470         int wrbuffer_refs, wake = 0;
1471
1472 retry:
1473         spin_lock(&inode->i_lock);
1474         if (ci->i_truncate_pending == 0) {
1475                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1476                 spin_unlock(&inode->i_lock);
1477                 return;
1478         }
1479
1480         /*
1481          * make sure any dirty snapped pages are flushed before we
1482          * possibly truncate them.. so write AND block!
1483          */
1484         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1485                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1486                      inode);
1487                 spin_unlock(&inode->i_lock);
1488                 filemap_write_and_wait_range(&inode->i_data, 0,
1489                                              inode->i_sb->s_maxbytes);
1490                 goto retry;
1491         }
1492
1493         to = ci->i_truncate_size;
1494         wrbuffer_refs = ci->i_wrbuffer_ref;
1495         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1496              ci->i_truncate_pending, to);
1497         spin_unlock(&inode->i_lock);
1498
1499         truncate_inode_pages(inode->i_mapping, to);
1500
1501         spin_lock(&inode->i_lock);
1502         ci->i_truncate_pending--;
1503         if (ci->i_truncate_pending == 0)
1504                 wake = 1;
1505         spin_unlock(&inode->i_lock);
1506
1507         if (wrbuffer_refs == 0)
1508                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1509         if (wake)
1510                 wake_up(&ci->i_cap_wq);
1511 }
1512
1513
1514 /*
1515  * symlinks
1516  */
1517 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1518 {
1519         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1520         nd_set_link(nd, ci->i_symlink);
1521         return NULL;
1522 }
1523
1524 static const struct inode_operations ceph_symlink_iops = {
1525         .readlink = generic_readlink,
1526         .follow_link = ceph_sym_follow_link,
1527 };
1528
1529 /*
1530  * setattr
1531  */
1532 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1533 {
1534         struct inode *inode = dentry->d_inode;
1535         struct ceph_inode_info *ci = ceph_inode(inode);
1536         struct inode *parent_inode = dentry->d_parent->d_inode;
1537         const unsigned int ia_valid = attr->ia_valid;
1538         struct ceph_mds_request *req;
1539         struct ceph_mds_client *mdsc = &ceph_sb_to_client(dentry->d_sb)->mdsc;
1540         int issued;
1541         int release = 0, dirtied = 0;
1542         int mask = 0;
1543         int err = 0;
1544
1545         if (ceph_snap(inode) != CEPH_NOSNAP)
1546                 return -EROFS;
1547
1548         __ceph_do_pending_vmtruncate(inode);
1549
1550         err = inode_change_ok(inode, attr);
1551         if (err != 0)
1552                 return err;
1553
1554         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1555                                        USE_AUTH_MDS);
1556         if (IS_ERR(req))
1557                 return PTR_ERR(req);
1558
1559         spin_lock(&inode->i_lock);
1560         issued = __ceph_caps_issued(ci, NULL);
1561         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1562
1563         if (ia_valid & ATTR_UID) {
1564                 dout("setattr %p uid %d -> %d\n", inode,
1565                      inode->i_uid, attr->ia_uid);
1566                 if (issued & CEPH_CAP_AUTH_EXCL) {
1567                         inode->i_uid = attr->ia_uid;
1568                         dirtied |= CEPH_CAP_AUTH_EXCL;
1569                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1570                            attr->ia_uid != inode->i_uid) {
1571                         req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1572                         mask |= CEPH_SETATTR_UID;
1573                         release |= CEPH_CAP_AUTH_SHARED;
1574                 }
1575         }
1576         if (ia_valid & ATTR_GID) {
1577                 dout("setattr %p gid %d -> %d\n", inode,
1578                      inode->i_gid, attr->ia_gid);
1579                 if (issued & CEPH_CAP_AUTH_EXCL) {
1580                         inode->i_gid = attr->ia_gid;
1581                         dirtied |= CEPH_CAP_AUTH_EXCL;
1582                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1583                            attr->ia_gid != inode->i_gid) {
1584                         req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1585                         mask |= CEPH_SETATTR_GID;
1586                         release |= CEPH_CAP_AUTH_SHARED;
1587                 }
1588         }
1589         if (ia_valid & ATTR_MODE) {
1590                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1591                      attr->ia_mode);
1592                 if (issued & CEPH_CAP_AUTH_EXCL) {
1593                         inode->i_mode = attr->ia_mode;
1594                         dirtied |= CEPH_CAP_AUTH_EXCL;
1595                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1596                            attr->ia_mode != inode->i_mode) {
1597                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1598                         mask |= CEPH_SETATTR_MODE;
1599                         release |= CEPH_CAP_AUTH_SHARED;
1600                 }
1601         }
1602
1603         if (ia_valid & ATTR_ATIME) {
1604                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1605                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1606                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1607                 if (issued & CEPH_CAP_FILE_EXCL) {
1608                         ci->i_time_warp_seq++;
1609                         inode->i_atime = attr->ia_atime;
1610                         dirtied |= CEPH_CAP_FILE_EXCL;
1611                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1612                            timespec_compare(&inode->i_atime,
1613                                             &attr->ia_atime) < 0) {
1614                         inode->i_atime = attr->ia_atime;
1615                         dirtied |= CEPH_CAP_FILE_WR;
1616                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1617                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1618                         ceph_encode_timespec(&req->r_args.setattr.atime,
1619                                              &attr->ia_atime);
1620                         mask |= CEPH_SETATTR_ATIME;
1621                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1622                                 CEPH_CAP_FILE_WR;
1623                 }
1624         }
1625         if (ia_valid & ATTR_MTIME) {
1626                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1627                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1628                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1629                 if (issued & CEPH_CAP_FILE_EXCL) {
1630                         ci->i_time_warp_seq++;
1631                         inode->i_mtime = attr->ia_mtime;
1632                         dirtied |= CEPH_CAP_FILE_EXCL;
1633                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1634                            timespec_compare(&inode->i_mtime,
1635                                             &attr->ia_mtime) < 0) {
1636                         inode->i_mtime = attr->ia_mtime;
1637                         dirtied |= CEPH_CAP_FILE_WR;
1638                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1639                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1640                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1641                                              &attr->ia_mtime);
1642                         mask |= CEPH_SETATTR_MTIME;
1643                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1644                                 CEPH_CAP_FILE_WR;
1645                 }
1646         }
1647         if (ia_valid & ATTR_SIZE) {
1648                 dout("setattr %p size %lld -> %lld\n", inode,
1649                      inode->i_size, attr->ia_size);
1650                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1651                         err = -EINVAL;
1652                         goto out;
1653                 }
1654                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1655                     attr->ia_size > inode->i_size) {
1656                         inode->i_size = attr->ia_size;
1657                         inode->i_blocks =
1658                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1659                         inode->i_ctime = attr->ia_ctime;
1660                         ci->i_reported_size = attr->ia_size;
1661                         dirtied |= CEPH_CAP_FILE_EXCL;
1662                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1663                            attr->ia_size != inode->i_size) {
1664                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1665                         req->r_args.setattr.old_size =
1666                                 cpu_to_le64(inode->i_size);
1667                         mask |= CEPH_SETATTR_SIZE;
1668                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1669                                 CEPH_CAP_FILE_WR;
1670                 }
1671         }
1672
1673         /* these do nothing */
1674         if (ia_valid & ATTR_CTIME) {
1675                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1676                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1677                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1678                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1679                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1680                      only ? "ctime only" : "ignored");
1681                 inode->i_ctime = attr->ia_ctime;
1682                 if (only) {
1683                         /*
1684                          * if kernel wants to dirty ctime but nothing else,
1685                          * we need to choose a cap to dirty under, or do
1686                          * a almost-no-op setattr
1687                          */
1688                         if (issued & CEPH_CAP_AUTH_EXCL)
1689                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1690                         else if (issued & CEPH_CAP_FILE_EXCL)
1691                                 dirtied |= CEPH_CAP_FILE_EXCL;
1692                         else if (issued & CEPH_CAP_XATTR_EXCL)
1693                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1694                         else
1695                                 mask |= CEPH_SETATTR_CTIME;
1696                 }
1697         }
1698         if (ia_valid & ATTR_FILE)
1699                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1700
1701         if (dirtied) {
1702                 __ceph_mark_dirty_caps(ci, dirtied);
1703                 inode->i_ctime = CURRENT_TIME;
1704         }
1705
1706         release &= issued;
1707         spin_unlock(&inode->i_lock);
1708
1709         if (mask) {
1710                 req->r_inode = igrab(inode);
1711                 req->r_inode_drop = release;
1712                 req->r_args.setattr.mask = cpu_to_le32(mask);
1713                 req->r_num_caps = 1;
1714                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1715         }
1716         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1717              ceph_cap_string(dirtied), mask);
1718
1719         ceph_mdsc_put_request(req);
1720         __ceph_do_pending_vmtruncate(inode);
1721         return err;
1722 out:
1723         spin_unlock(&inode->i_lock);
1724         ceph_mdsc_put_request(req);
1725         return err;
1726 }
1727
1728 /*
1729  * Verify that we have a lease on the given mask.  If not,
1730  * do a getattr against an mds.
1731  */
1732 int ceph_do_getattr(struct inode *inode, int mask)
1733 {
1734         struct ceph_client *client = ceph_sb_to_client(inode->i_sb);
1735         struct ceph_mds_client *mdsc = &client->mdsc;
1736         struct ceph_mds_request *req;
1737         int err;
1738
1739         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1740                 dout("do_getattr inode %p SNAPDIR\n", inode);
1741                 return 0;
1742         }
1743
1744         dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask));
1745         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1746                 return 0;
1747
1748         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1749         if (IS_ERR(req))
1750                 return PTR_ERR(req);
1751         req->r_inode = igrab(inode);
1752         req->r_num_caps = 1;
1753         req->r_args.getattr.mask = cpu_to_le32(mask);
1754         err = ceph_mdsc_do_request(mdsc, NULL, req);
1755         ceph_mdsc_put_request(req);
1756         dout("do_getattr result=%d\n", err);
1757         return err;
1758 }
1759
1760
1761 /*
1762  * Check inode permissions.  We verify we have a valid value for
1763  * the AUTH cap, then call the generic handler.
1764  */
1765 int ceph_permission(struct inode *inode, int mask)
1766 {
1767         int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1768
1769         if (!err)
1770                 err = generic_permission(inode, mask, NULL);
1771         return err;
1772 }
1773
1774 /*
1775  * Get all attributes.  Hopefully somedata we'll have a statlite()
1776  * and can limit the fields we require to be accurate.
1777  */
1778 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1779                  struct kstat *stat)
1780 {
1781         struct inode *inode = dentry->d_inode;
1782         struct ceph_inode_info *ci = ceph_inode(inode);
1783         int err;
1784
1785         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1786         if (!err) {
1787                 generic_fillattr(inode, stat);
1788                 stat->ino = inode->i_ino;
1789                 if (ceph_snap(inode) != CEPH_NOSNAP)
1790                         stat->dev = ceph_snap(inode);
1791                 else
1792                         stat->dev = 0;
1793                 if (S_ISDIR(inode->i_mode)) {
1794                         stat->size = ci->i_rbytes;
1795                         stat->blocks = 0;
1796                         stat->blksize = 65536;
1797                 }
1798         }
1799         return err;
1800 }