fs/ceph/inode.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/slab.h>
   6 #include <linux/string.h>
   7 #include <linux/uaccess.h>
   8 #include <linux/kernel.h>
   9 #include <linux/writeback.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/posix_acl.h>
  12 #include <linux/random.h>
  13
  14 #include "super.h"
  15 #include "mds_client.h"
  16 #include "cache.h"
  17 #include <linux/ceph/decode.h>
  18
  19 /*
  20  * Ceph inode operations
  21  *
  22  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  23  * setattr, etc.), xattr helpers, and helpers for assimilating
  24  * metadata returned by the MDS into our cache.
  25  *
  26  * Also define helpers for doing asynchronous writeback, invalidation,
  27  * and truncation for the benefit of those who can't afford to block
  28  * (typically because they are in the message handler path).
  29  */
  30
  31 static const struct inode_operations ceph_symlink_iops;
  32
  33 static void ceph_invalidate_work(struct work_struct *work);
  34 static void ceph_writeback_work(struct work_struct *work);
  35 static void ceph_vmtruncate_work(struct work_struct *work);
  36
  37 /*
  38  * find or create an inode, given the ceph ino number
  39  */
  40 static int ceph_set_ino_cb(struct inode *inode, void *data)
  41 {
  42         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  43         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  44         return 0;
  45 }
  46
  47 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  48 {
  49         struct inode *inode;
  50         ino_t t = ceph_vino_to_ino(vino);
  51
  52         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  53         if (inode == NULL)
  54                 return ERR_PTR(-ENOMEM);
  55         if (inode->i_state & I_NEW) {
  56                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  57                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  58                 unlock_new_inode(inode);
  59         }
  60
  61         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  62              vino.snap, inode);
  63         return inode;
  64 }
  65
  66 /*
  67  * get/constuct snapdir inode for a given directory
  68  */
  69 struct inode *ceph_get_snapdir(struct inode *parent)
  70 {
  71         struct ceph_vino vino = {
  72                 .ino = ceph_ino(parent),
  73                 .snap = CEPH_SNAPDIR,
  74         };
  75         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  76         struct ceph_inode_info *ci = ceph_inode(inode);
  77
  78         BUG_ON(!S_ISDIR(parent->i_mode));
  79         if (IS_ERR(inode))
  80                 return inode;
  81         inode->i_mode = parent->i_mode;
  82         inode->i_uid = parent->i_uid;
  83         inode->i_gid = parent->i_gid;
  84         inode->i_op = &ceph_snapdir_iops;
  85         inode->i_fop = &ceph_snapdir_fops;
  86         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  87         ci->i_rbytes = 0;
  88         return inode;
  89 }
  90
  91 const struct inode_operations ceph_file_iops = {
  92         .permission = ceph_permission,
  93         .setattr = ceph_setattr,
  94         .getattr = ceph_getattr,
  95         .setxattr = ceph_setxattr,
  96         .getxattr = ceph_getxattr,
  97         .listxattr = ceph_listxattr,
  98         .removexattr = ceph_removexattr,
  99         .get_acl = ceph_get_acl,
 100         .set_acl = ceph_set_acl,
 101 };
 102
 103
 104 /*
 105  * We use a 'frag tree' to keep track of the MDS's directory fragments
 106  * for a given inode (usually there is just a single fragment).  We
 107  * need to know when a child frag is delegated to a new MDS, or when
 108  * it is flagged as replicated, so we can direct our requests
 109  * accordingly.
 110  */
 111
 112 /*
 113  * find/create a frag in the tree
 114  */
 115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 116                                                     u32 f)
 117 {
 118         struct rb_node **p;
 119         struct rb_node *parent = NULL;
 120         struct ceph_inode_frag *frag;
 121         int c;
 122
 123         p = &ci->i_fragtree.rb_node;
 124         while (*p) {
 125                 parent = *p;
 126                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 127                 c = ceph_frag_compare(f, frag->frag);
 128                 if (c < 0)
 129                         p = &(*p)->rb_left;
 130                 else if (c > 0)
 131                         p = &(*p)->rb_right;
 132                 else
 133                         return frag;
 134         }
 135
 136         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 137         if (!frag) {
 138                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 139                        "frag %x\n", &ci->vfs_inode,
 140                        ceph_vinop(&ci->vfs_inode), f);
 141                 return ERR_PTR(-ENOMEM);
 142         }
 143         frag->frag = f;
 144         frag->split_by = 0;
 145         frag->mds = -1;
 146         frag->ndist = 0;
 147
 148         rb_link_node(&frag->node, parent, p);
 149         rb_insert_color(&frag->node, &ci->i_fragtree);
 150
 151         dout("get_or_create_frag added %llx.%llx frag %x\n",
 152              ceph_vinop(&ci->vfs_inode), f);
 153         return frag;
 154 }
 155
 156 /*
 157  * find a specific frag @f
 158  */
 159 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 160 {
 161         struct rb_node *n = ci->i_fragtree.rb_node;
 162
 163         while (n) {
 164                 struct ceph_inode_frag *frag =
 165                         rb_entry(n, struct ceph_inode_frag, node);
 166                 int c = ceph_frag_compare(f, frag->frag);
 167                 if (c < 0)
 168                         n = n->rb_left;
 169                 else if (c > 0)
 170                         n = n->rb_right;
 171                 else
 172                         return frag;
 173         }
 174         return NULL;
 175 }
 176
 177 /*
 178  * Choose frag containing the given value @v.  If @pfrag is
 179  * specified, copy the frag delegation info to the caller if
 180  * it is present.
 181  */
 182 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 183                               struct ceph_inode_frag *pfrag, int *found)
 184 {
 185         u32 t = ceph_frag_make(0, 0);
 186         struct ceph_inode_frag *frag;
 187         unsigned nway, i;
 188         u32 n;
 189
 190         if (found)
 191                 *found = 0;
 192
 193         while (1) {
 194                 WARN_ON(!ceph_frag_contains_value(t, v));
 195                 frag = __ceph_find_frag(ci, t);
 196                 if (!frag)
 197                         break; /* t is a leaf */
 198                 if (frag->split_by == 0) {
 199                         if (pfrag)
 200                                 memcpy(pfrag, frag, sizeof(*pfrag));
 201                         if (found)
 202                                 *found = 1;
 203                         break;
 204                 }
 205
 206                 /* choose child */
 207                 nway = 1 << frag->split_by;
 208                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 209                      frag->split_by, nway);
 210                 for (i = 0; i < nway; i++) {
 211                         n = ceph_frag_make_child(t, frag->split_by, i);
 212                         if (ceph_frag_contains_value(n, v)) {
 213                                 t = n;
 214                                 break;
 215                         }
 216                 }
 217                 BUG_ON(i == nway);
 218         }
 219         dout("choose_frag(%x) = %x\n", v, t);
 220
 221         return t;
 222 }
 223
 224 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 225                      struct ceph_inode_frag *pfrag, int *found)
 226 {
 227         u32 ret;
 228         mutex_lock(&ci->i_fragtree_mutex);
 229         ret = __ceph_choose_frag(ci, v, pfrag, found);
 230         mutex_unlock(&ci->i_fragtree_mutex);
 231         return ret;
 232 }
 233
 234 /*
 235  * Process dirfrag (delegation) info from the mds.  Include leaf
 236  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 237  * branches/splits are included in i_fragtree)
 238  */
 239 static int ceph_fill_dirfrag(struct inode *inode,
 240                              struct ceph_mds_reply_dirfrag *dirinfo)
 241 {
 242         struct ceph_inode_info *ci = ceph_inode(inode);
 243         struct ceph_inode_frag *frag;
 244         u32 id = le32_to_cpu(dirinfo->frag);
 245         int mds = le32_to_cpu(dirinfo->auth);
 246         int ndist = le32_to_cpu(dirinfo->ndist);
 247         int diri_auth = -1;
 248         int i;
 249         int err = 0;
 250
 251         spin_lock(&ci->i_ceph_lock);
 252         if (ci->i_auth_cap)
 253                 diri_auth = ci->i_auth_cap->mds;
 254         spin_unlock(&ci->i_ceph_lock);
 255
 256         mutex_lock(&ci->i_fragtree_mutex);
 257         if (ndist == 0 && mds == diri_auth) {
 258                 /* no delegation info needed. */
 259                 frag = __ceph_find_frag(ci, id);
 260                 if (!frag)
 261                         goto out;
 262                 if (frag->split_by == 0) {
 263                         /* tree leaf, remove */
 264                         dout("fill_dirfrag removed %llx.%llx frag %x"
 265                              " (no ref)\n", ceph_vinop(inode), id);
 266                         rb_erase(&frag->node, &ci->i_fragtree);
 267                         kfree(frag);
 268                 } else {
 269                         /* tree branch, keep and clear */
 270                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 271                              " referral\n", ceph_vinop(inode), id);
 272                         frag->mds = -1;
 273                         frag->ndist = 0;
 274                 }
 275                 goto out;
 276         }
 277
 278
 279         /* find/add this frag to store mds delegation info */
 280         frag = __get_or_create_frag(ci, id);
 281         if (IS_ERR(frag)) {
 282                 /* this is not the end of the world; we can continue
 283                    with bad/inaccurate delegation info */
 284                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 285                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 286                 err = -ENOMEM;
 287                 goto out;
 288         }
 289
 290         frag->mds = mds;
 291         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 292         for (i = 0; i < frag->ndist; i++)
 293                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 294         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 295              ceph_vinop(inode), frag->frag, frag->ndist);
 296
 297 out:
 298         mutex_unlock(&ci->i_fragtree_mutex);
 299         return err;
 300 }
 301
 302 static int ceph_fill_fragtree(struct inode *inode,
 303                               struct ceph_frag_tree_head *fragtree,
 304                               struct ceph_mds_reply_dirfrag *dirinfo)
 305 {
 306         struct ceph_inode_info *ci = ceph_inode(inode);
 307         struct ceph_inode_frag *frag;
 308         struct rb_node *rb_node;
 309         int i;
 310         u32 id, nsplits;
 311         bool update = false;
 312
 313         mutex_lock(&ci->i_fragtree_mutex);
 314         nsplits = le32_to_cpu(fragtree->nsplits);
 315         if (nsplits) {
 316                 i = prandom_u32() % nsplits;
 317                 id = le32_to_cpu(fragtree->splits[i].frag);
 318                 if (!__ceph_find_frag(ci, id))
 319                         update = true;
 320         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 321                 rb_node = rb_first(&ci->i_fragtree);
 322                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 323                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 324                         update = true;
 325         }
 326         if (!update && dirinfo) {
 327                 id = le32_to_cpu(dirinfo->frag);
 328                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 329                         update = true;
 330         }
 331         if (!update)
 332                 goto out_unlock;
 333
 334         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 335         rb_node = rb_first(&ci->i_fragtree);
 336         for (i = 0; i < nsplits; i++) {
 337                 id = le32_to_cpu(fragtree->splits[i].frag);
 338                 frag = NULL;
 339                 while (rb_node) {
 340                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 341                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 342                                 if (frag->frag != id)
 343                                         frag = NULL;
 344                                 else
 345                                         rb_node = rb_next(rb_node);
 346                                 break;
 347                         }
 348                         rb_node = rb_next(rb_node);
 349                         rb_erase(&frag->node, &ci->i_fragtree);
 350                         kfree(frag);
 351                         frag = NULL;
 352                 }
 353                 if (!frag) {
 354                         frag = __get_or_create_frag(ci, id);
 355                         if (IS_ERR(frag))
 356                                 continue;
 357                 }
 358                 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
 359                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 360         }
 361         while (rb_node) {
 362                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 363                 rb_node = rb_next(rb_node);
 364                 rb_erase(&frag->node, &ci->i_fragtree);
 365                 kfree(frag);
 366         }
 367 out_unlock:
 368         mutex_unlock(&ci->i_fragtree_mutex);
 369         return 0;
 370 }
 371
 372 /*
 373  * initialize a newly allocated inode.
 374  */
 375 struct inode *ceph_alloc_inode(struct super_block *sb)
 376 {
 377         struct ceph_inode_info *ci;
 378         int i;
 379
 380         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 381         if (!ci)
 382                 return NULL;
 383
 384         dout("alloc_inode %p\n", &ci->vfs_inode);
 385
 386         spin_lock_init(&ci->i_ceph_lock);
 387
 388         ci->i_version = 0;
 389         ci->i_inline_version = 0;
 390         ci->i_time_warp_seq = 0;
 391         ci->i_ceph_flags = 0;
 392         ci->i_ordered_count = 0;
 393         atomic_set(&ci->i_release_count, 1);
 394         atomic_set(&ci->i_complete_count, 0);
 395         ci->i_symlink = NULL;
 396
 397         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 398
 399         ci->i_fragtree = RB_ROOT;
 400         mutex_init(&ci->i_fragtree_mutex);
 401
 402         ci->i_xattrs.blob = NULL;
 403         ci->i_xattrs.prealloc_blob = NULL;
 404         ci->i_xattrs.dirty = false;
 405         ci->i_xattrs.index = RB_ROOT;
 406         ci->i_xattrs.count = 0;
 407         ci->i_xattrs.names_size = 0;
 408         ci->i_xattrs.vals_size = 0;
 409         ci->i_xattrs.version = 0;
 410         ci->i_xattrs.index_version = 0;
 411
 412         ci->i_caps = RB_ROOT;
 413         ci->i_auth_cap = NULL;
 414         ci->i_dirty_caps = 0;
 415         ci->i_flushing_caps = 0;
 416         INIT_LIST_HEAD(&ci->i_dirty_item);
 417         INIT_LIST_HEAD(&ci->i_flushing_item);
 418         ci->i_cap_flush_seq = 0;
 419         ci->i_cap_flush_last_tid = 0;
 420         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 421         init_waitqueue_head(&ci->i_cap_wq);
 422         ci->i_hold_caps_min = 0;
 423         ci->i_hold_caps_max = 0;
 424         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 425         INIT_LIST_HEAD(&ci->i_cap_snaps);
 426         ci->i_head_snapc = NULL;
 427         ci->i_snap_caps = 0;
 428
 429         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 430                 ci->i_nr_by_mode[i] = 0;
 431
 432         mutex_init(&ci->i_truncate_mutex);
 433         ci->i_truncate_seq = 0;
 434         ci->i_truncate_size = 0;
 435         ci->i_truncate_pending = 0;
 436
 437         ci->i_max_size = 0;
 438         ci->i_reported_size = 0;
 439         ci->i_wanted_max_size = 0;
 440         ci->i_requested_max_size = 0;
 441
 442         ci->i_pin_ref = 0;
 443         ci->i_rd_ref = 0;
 444         ci->i_rdcache_ref = 0;
 445         ci->i_wr_ref = 0;
 446         ci->i_wb_ref = 0;
 447         ci->i_wrbuffer_ref = 0;
 448         ci->i_wrbuffer_ref_head = 0;
 449         ci->i_shared_gen = 0;
 450         ci->i_rdcache_gen = 0;
 451         ci->i_rdcache_revoking = 0;
 452
 453         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 454         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 455         spin_lock_init(&ci->i_unsafe_lock);
 456
 457         ci->i_snap_realm = NULL;
 458         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 459         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 460
 461         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 462         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 463
 464         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 465
 466         ceph_fscache_inode_init(ci);
 467
 468         return &ci->vfs_inode;
 469 }
 470
 471 static void ceph_i_callback(struct rcu_head *head)
 472 {
 473         struct inode *inode = container_of(head, struct inode, i_rcu);
 474         struct ceph_inode_info *ci = ceph_inode(inode);
 475
 476         kmem_cache_free(ceph_inode_cachep, ci);
 477 }
 478
 479 void ceph_destroy_inode(struct inode *inode)
 480 {
 481         struct ceph_inode_info *ci = ceph_inode(inode);
 482         struct ceph_inode_frag *frag;
 483         struct rb_node *n;
 484
 485         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 486
 487         ceph_fscache_unregister_inode_cookie(ci);
 488
 489         ceph_queue_caps_release(inode);
 490
 491         /*
 492          * we may still have a snap_realm reference if there are stray
 493          * caps in i_snap_caps.
 494          */
 495         if (ci->i_snap_realm) {
 496                 struct ceph_mds_client *mdsc =
 497                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 498                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 499
 500                 dout(" dropping residual ref to snap realm %p\n", realm);
 501                 spin_lock(&realm->inodes_with_caps_lock);
 502                 list_del_init(&ci->i_snap_realm_item);
 503                 spin_unlock(&realm->inodes_with_caps_lock);
 504                 ceph_put_snap_realm(mdsc, realm);
 505         }
 506
 507         kfree(ci->i_symlink);
 508         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 509                 frag = rb_entry(n, struct ceph_inode_frag, node);
 510                 rb_erase(n, &ci->i_fragtree);
 511                 kfree(frag);
 512         }
 513
 514         __ceph_destroy_xattrs(ci);
 515         if (ci->i_xattrs.blob)
 516                 ceph_buffer_put(ci->i_xattrs.blob);
 517         if (ci->i_xattrs.prealloc_blob)
 518                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 519
 520         call_rcu(&inode->i_rcu, ceph_i_callback);
 521 }
 522
 523 int ceph_drop_inode(struct inode *inode)
 524 {
 525         /*
 526          * Positve dentry and corresponding inode are always accompanied
 527          * in MDS reply. So no need to keep inode in the cache after
 528          * dropping all its aliases.
 529          */
 530         return 1;
 531 }
 532
 533 /*
 534  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 535  * careful because either the client or MDS may have more up to date
 536  * info, depending on which capabilities are held, and whether
 537  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 538  * and size are monotonically increasing, except when utimes() or
 539  * truncate() increments the corresponding _seq values.)
 540  */
 541 int ceph_fill_file_size(struct inode *inode, int issued,
 542                         u32 truncate_seq, u64 truncate_size, u64 size)
 543 {
 544         struct ceph_inode_info *ci = ceph_inode(inode);
 545         int queue_trunc = 0;
 546
 547         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 548             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 549                 dout("size %lld -> %llu\n", inode->i_size, size);
 550                 inode->i_size = size;
 551                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 552                 ci->i_reported_size = size;
 553                 if (truncate_seq != ci->i_truncate_seq) {
 554                         dout("truncate_seq %u -> %u\n",
 555                              ci->i_truncate_seq, truncate_seq);
 556                         ci->i_truncate_seq = truncate_seq;
 557
 558                         /* the MDS should have revoked these caps */
 559                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 560                                                CEPH_CAP_FILE_RD |
 561                                                CEPH_CAP_FILE_WR |
 562                                                CEPH_CAP_FILE_LAZYIO));
 563                         /*
 564                          * If we hold relevant caps, or in the case where we're
 565                          * not the only client referencing this file and we
 566                          * don't hold those caps, then we need to check whether
 567                          * the file is either opened or mmaped
 568                          */
 569                         if ((issued & (CEPH_CAP_FILE_CACHE|
 570                                        CEPH_CAP_FILE_BUFFER)) ||
 571                             mapping_mapped(inode->i_mapping) ||
 572                             __ceph_caps_file_wanted(ci)) {
 573                                 ci->i_truncate_pending++;
 574                                 queue_trunc = 1;
 575                         }
 576                 }
 577         }
 578         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 579             ci->i_truncate_size != truncate_size) {
 580                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 581                      truncate_size);
 582                 ci->i_truncate_size = truncate_size;
 583         }
 584
 585         if (queue_trunc)
 586                 ceph_fscache_invalidate(inode);
 587
 588         return queue_trunc;
 589 }
 590
 591 void ceph_fill_file_time(struct inode *inode, int issued,
 592                          u64 time_warp_seq, struct timespec *ctime,
 593                          struct timespec *mtime, struct timespec *atime)
 594 {
 595         struct ceph_inode_info *ci = ceph_inode(inode);
 596         int warn = 0;
 597
 598         if (issued & (CEPH_CAP_FILE_EXCL|
 599                       CEPH_CAP_FILE_WR|
 600                       CEPH_CAP_FILE_BUFFER|
 601                       CEPH_CAP_AUTH_EXCL|
 602                       CEPH_CAP_XATTR_EXCL)) {
 603                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 604                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 605                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 606                              ctime->tv_sec, ctime->tv_nsec);
 607                         inode->i_ctime = *ctime;
 608                 }
 609                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 610                         /* the MDS did a utimes() */
 611                         dout("mtime %ld.%09ld -> %ld.%09ld "
 612                              "tw %d -> %d\n",
 613                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 614                              mtime->tv_sec, mtime->tv_nsec,
 615                              ci->i_time_warp_seq, (int)time_warp_seq);
 616
 617                         inode->i_mtime = *mtime;
 618                         inode->i_atime = *atime;
 619                         ci->i_time_warp_seq = time_warp_seq;
 620                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 621                         /* nobody did utimes(); take the max */
 622                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 623                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 624                                      inode->i_mtime.tv_sec,
 625                                      inode->i_mtime.tv_nsec,
 626                                      mtime->tv_sec, mtime->tv_nsec);
 627                                 inode->i_mtime = *mtime;
 628                         }
 629                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 630                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 631                                      inode->i_atime.tv_sec,
 632                                      inode->i_atime.tv_nsec,
 633                                      atime->tv_sec, atime->tv_nsec);
 634                                 inode->i_atime = *atime;
 635                         }
 636                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 637                         /* we did a utimes(); ignore mds values */
 638                 } else {
 639                         warn = 1;
 640                 }
 641         } else {
 642                 /* we have no write|excl caps; whatever the MDS says is true */
 643                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 644                         inode->i_ctime = *ctime;
 645                         inode->i_mtime = *mtime;
 646                         inode->i_atime = *atime;
 647                         ci->i_time_warp_seq = time_warp_seq;
 648                 } else {
 649                         warn = 1;
 650                 }
 651         }
 652         if (warn) /* time_warp_seq shouldn't go backwards */
 653                 dout("%p mds time_warp_seq %llu < %u\n",
 654                      inode, time_warp_seq, ci->i_time_warp_seq);
 655 }
 656
 657 /*
 658  * Populate an inode based on info from mds.  May be called on new or
 659  * existing inodes.
 660  */
 661 static int fill_inode(struct inode *inode, struct page *locked_page,
 662                       struct ceph_mds_reply_info_in *iinfo,
 663                       struct ceph_mds_reply_dirfrag *dirinfo,
 664                       struct ceph_mds_session *session,
 665                       unsigned long ttl_from, int cap_fmode,
 666                       struct ceph_cap_reservation *caps_reservation)
 667 {
 668         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 669         struct ceph_mds_reply_inode *info = iinfo->in;
 670         struct ceph_inode_info *ci = ceph_inode(inode);
 671         int issued = 0, implemented, new_issued;
 672         struct timespec mtime, atime, ctime;
 673         struct ceph_buffer *xattr_blob = NULL;
 674         struct ceph_cap *new_cap = NULL;
 675         int err = 0;
 676         bool wake = false;
 677         bool queue_trunc = false;
 678         bool new_version = false;
 679         bool fill_inline = false;
 680
 681         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 682              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 683              ci->i_version);
 684
 685         /* prealloc new cap struct */
 686         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 687                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 688
 689         /*
 690          * prealloc xattr data, if it looks like we'll need it.  only
 691          * if len > 4 (meaning there are actually xattrs; the first 4
 692          * bytes are the xattr count).
 693          */
 694         if (iinfo->xattr_len > 4) {
 695                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 696                 if (!xattr_blob)
 697                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 698                                iinfo->xattr_len);
 699         }
 700
 701         spin_lock(&ci->i_ceph_lock);
 702
 703         /*
 704          * provided version will be odd if inode value is projected,
 705          * even if stable.  skip the update if we have newer stable
 706          * info (ours>=theirs, e.g. due to racing mds replies), unless
 707          * we are getting projected (unstable) info (in which case the
 708          * version is odd, and we want ours>theirs).
 709          *   us   them
 710          *   2    2     skip
 711          *   3    2     skip
 712          *   3    3     update
 713          */
 714         if (ci->i_version == 0 ||
 715             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 716              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 717                 new_version = true;
 718
 719         issued = __ceph_caps_issued(ci, &implemented);
 720         issued |= implemented | __ceph_caps_dirty(ci);
 721         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 722
 723         /* update inode */
 724         ci->i_version = le64_to_cpu(info->version);
 725         inode->i_version++;
 726         inode->i_rdev = le32_to_cpu(info->rdev);
 727         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 728
 729         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 730             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 731                 inode->i_mode = le32_to_cpu(info->mode);
 732                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 733                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 734                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 735                      from_kuid(&init_user_ns, inode->i_uid),
 736                      from_kgid(&init_user_ns, inode->i_gid));
 737         }
 738
 739         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 740             (issued & CEPH_CAP_LINK_EXCL) == 0)
 741                 set_nlink(inode, le32_to_cpu(info->nlink));
 742
 743         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 744                 /* be careful with mtime, atime, size */
 745                 ceph_decode_timespec(&atime, &info->atime);
 746                 ceph_decode_timespec(&mtime, &info->mtime);
 747                 ceph_decode_timespec(&ctime, &info->ctime);
 748                 ceph_fill_file_time(inode, issued,
 749                                 le32_to_cpu(info->time_warp_seq),
 750                                 &ctime, &mtime, &atime);
 751         }
 752
 753         if (new_version ||
 754             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 755                 ci->i_layout = info->layout;
 756                 queue_trunc = ceph_fill_file_size(inode, issued,
 757                                         le32_to_cpu(info->truncate_seq),
 758                                         le64_to_cpu(info->truncate_size),
 759                                         le64_to_cpu(info->size));
 760                 /* only update max_size on auth cap */
 761                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 762                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 763                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 764                                         le64_to_cpu(info->max_size));
 765                         ci->i_max_size = le64_to_cpu(info->max_size);
 766                 }
 767         }
 768
 769         /* xattrs */
 770         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 771         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 772             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 773                 if (ci->i_xattrs.blob)
 774                         ceph_buffer_put(ci->i_xattrs.blob);
 775                 ci->i_xattrs.blob = xattr_blob;
 776                 if (xattr_blob)
 777                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 778                                iinfo->xattr_data, iinfo->xattr_len);
 779                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 780                 ceph_forget_all_cached_acls(inode);
 781                 xattr_blob = NULL;
 782         }
 783
 784         inode->i_mapping->a_ops = &ceph_aops;
 785
 786         switch (inode->i_mode & S_IFMT) {
 787         case S_IFIFO:
 788         case S_IFBLK:
 789         case S_IFCHR:
 790         case S_IFSOCK:
 791                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 792                 inode->i_op = &ceph_file_iops;
 793                 break;
 794         case S_IFREG:
 795                 inode->i_op = &ceph_file_iops;
 796                 inode->i_fop = &ceph_file_fops;
 797                 break;
 798         case S_IFLNK:
 799                 inode->i_op = &ceph_symlink_iops;
 800                 if (!ci->i_symlink) {
 801                         u32 symlen = iinfo->symlink_len;
 802                         char *sym;
 803
 804                         spin_unlock(&ci->i_ceph_lock);
 805
 806                         err = -EINVAL;
 807                         if (WARN_ON(symlen != inode->i_size))
 808                                 goto out;
 809
 810                         err = -ENOMEM;
 811                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 812                         if (!sym)
 813                                 goto out;
 814
 815                         spin_lock(&ci->i_ceph_lock);
 816                         if (!ci->i_symlink)
 817                                 ci->i_symlink = sym;
 818                         else
 819                                 kfree(sym); /* lost a race */
 820                 }
 821                 inode->i_link = ci->i_symlink;
 822                 break;
 823         case S_IFDIR:
 824                 inode->i_op = &ceph_dir_iops;
 825                 inode->i_fop = &ceph_dir_fops;
 826
 827                 ci->i_dir_layout = iinfo->dir_layout;
 828
 829                 ci->i_files = le64_to_cpu(info->files);
 830                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 831                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 832                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 833                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 834                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 835                 break;
 836         default:
 837                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 838                        ceph_vinop(inode), inode->i_mode);
 839         }
 840
 841         /* were we issued a capability? */
 842         if (info->cap.caps) {
 843                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 844                         unsigned caps = le32_to_cpu(info->cap.caps);
 845                         ceph_add_cap(inode, session,
 846                                      le64_to_cpu(info->cap.cap_id),
 847                                      cap_fmode, caps,
 848                                      le32_to_cpu(info->cap.wanted),
 849                                      le32_to_cpu(info->cap.seq),
 850                                      le32_to_cpu(info->cap.mseq),
 851                                      le64_to_cpu(info->cap.realm),
 852                                      info->cap.flags, &new_cap);
 853
 854                         /* set dir completion flag? */
 855                         if (S_ISDIR(inode->i_mode) &&
 856                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 857                             (caps & CEPH_CAP_FILE_SHARED) &&
 858                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 859                             !__ceph_dir_is_complete(ci)) {
 860                                 dout(" marking %p complete (empty)\n", inode);
 861                                 __ceph_dir_set_complete(ci,
 862                                         atomic_read(&ci->i_release_count),
 863                                         ci->i_ordered_count);
 864                         }
 865
 866                         wake = true;
 867                 } else {
 868                         dout(" %p got snap_caps %s\n", inode,
 869                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 870                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 871                         if (cap_fmode >= 0)
 872                                 __ceph_get_fmode(ci, cap_fmode);
 873                 }
 874         } else if (cap_fmode >= 0) {
 875                 pr_warn("mds issued no caps on %llx.%llx\n",
 876                            ceph_vinop(inode));
 877                 __ceph_get_fmode(ci, cap_fmode);
 878         }
 879
 880         if (iinfo->inline_version > 0 &&
 881             iinfo->inline_version >= ci->i_inline_version) {
 882                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 883                 ci->i_inline_version = iinfo->inline_version;
 884                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 885                     (locked_page ||
 886                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 887                         fill_inline = true;
 888         }
 889
 890         spin_unlock(&ci->i_ceph_lock);
 891
 892         if (fill_inline)
 893                 ceph_fill_inline_data(inode, locked_page,
 894                                       iinfo->inline_data, iinfo->inline_len);
 895
 896         if (wake)
 897                 wake_up_all(&ci->i_cap_wq);
 898
 899         /* queue truncate if we saw i_size decrease */
 900         if (queue_trunc)
 901                 ceph_queue_vmtruncate(inode);
 902
 903         /* populate frag tree */
 904         if (S_ISDIR(inode->i_mode))
 905                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 906
 907         /* update delegation info? */
 908         if (dirinfo)
 909                 ceph_fill_dirfrag(inode, dirinfo);
 910
 911         err = 0;
 912 out:
 913         if (new_cap)
 914                 ceph_put_cap(mdsc, new_cap);
 915         if (xattr_blob)
 916                 ceph_buffer_put(xattr_blob);
 917         return err;
 918 }
 919
 920 /*
 921  * caller should hold session s_mutex.
 922  */
 923 static void update_dentry_lease(struct dentry *dentry,
 924                                 struct ceph_mds_reply_lease *lease,
 925                                 struct ceph_mds_session *session,
 926                                 unsigned long from_time)
 927 {
 928         struct ceph_dentry_info *di = ceph_dentry(dentry);
 929         long unsigned duration = le32_to_cpu(lease->duration_ms);
 930         long unsigned ttl = from_time + (duration * HZ) / 1000;
 931         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 932         struct inode *dir;
 933
 934         /* only track leases on regular dentries */
 935         if (dentry->d_op != &ceph_dentry_ops)
 936                 return;
 937
 938         spin_lock(&dentry->d_lock);
 939         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 940              dentry, duration, ttl);
 941
 942         /* make lease_rdcache_gen match directory */
 943         dir = d_inode(dentry->d_parent);
 944         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 945
 946         if (duration == 0)
 947                 goto out_unlock;
 948
 949         if (di->lease_gen == session->s_cap_gen &&
 950             time_before(ttl, dentry->d_time))
 951                 goto out_unlock;  /* we already have a newer lease. */
 952
 953         if (di->lease_session && di->lease_session != session)
 954                 goto out_unlock;
 955
 956         ceph_dentry_lru_touch(dentry);
 957
 958         if (!di->lease_session)
 959                 di->lease_session = ceph_get_mds_session(session);
 960         di->lease_gen = session->s_cap_gen;
 961         di->lease_seq = le32_to_cpu(lease->seq);
 962         di->lease_renew_after = half_ttl;
 963         di->lease_renew_from = 0;
 964         dentry->d_time = ttl;
 965 out_unlock:
 966         spin_unlock(&dentry->d_lock);
 967         return;
 968 }
 969
 970 /*
 971  * splice a dentry to an inode.
 972  * caller must hold directory i_mutex for this to be safe.
 973  *
 974  * we will only rehash the resulting dentry if @prehash is
 975  * true; @prehash will be set to false (for the benefit of
 976  * the caller) if we fail.
 977  */
 978 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 979                                     bool *prehash)
 980 {
 981         struct dentry *realdn;
 982
 983         BUG_ON(d_inode(dn));
 984
 985         /* dn must be unhashed */
 986         if (!d_unhashed(dn))
 987                 d_drop(dn);
 988         realdn = d_splice_alias(in, dn);
 989         if (IS_ERR(realdn)) {
 990                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 991                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 992                 if (prehash)
 993                         *prehash = false; /* don't rehash on error */
 994                 dn = realdn; /* note realdn contains the error */
 995                 goto out;
 996         } else if (realdn) {
 997                 dout("dn %p (%d) spliced with %p (%d) "
 998                      "inode %p ino %llx.%llx\n",
 999                      dn, d_count(dn),
1000                      realdn, d_count(realdn),
1001                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1002                 dput(dn);
1003                 dn = realdn;
1004         } else {
1005                 BUG_ON(!ceph_dentry(dn));
1006                 dout("dn %p attached to %p ino %llx.%llx\n",
1007                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1008         }
1009         if ((!prehash || *prehash) && d_unhashed(dn))
1010                 d_rehash(dn);
1011 out:
1012         return dn;
1013 }
1014
1015 /*
1016  * Incorporate results into the local cache.  This is either just
1017  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1018  * after a lookup).
1019  *
1020  * A reply may contain
1021  *         a directory inode along with a dentry.
1022  *  and/or a target inode
1023  *
1024  * Called with snap_rwsem (read).
1025  */
1026 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1027                     struct ceph_mds_session *session)
1028 {
1029         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1030         struct inode *in = NULL;
1031         struct ceph_vino vino;
1032         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1033         int err = 0;
1034
1035         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1036              rinfo->head->is_dentry, rinfo->head->is_target);
1037
1038 #if 0
1039         /*
1040          * Debugging hook:
1041          *
1042          * If we resend completed ops to a recovering mds, we get no
1043          * trace.  Since that is very rare, pretend this is the case
1044          * to ensure the 'no trace' handlers in the callers behave.
1045          *
1046          * Fill in inodes unconditionally to avoid breaking cap
1047          * invariants.
1048          */
1049         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1050                 pr_info("fill_trace faking empty trace on %lld %s\n",
1051                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1052                 if (rinfo->head->is_dentry) {
1053                         rinfo->head->is_dentry = 0;
1054                         err = fill_inode(req->r_locked_dir,
1055                                          &rinfo->diri, rinfo->dirfrag,
1056                                          session, req->r_request_started, -1);
1057                 }
1058                 if (rinfo->head->is_target) {
1059                         rinfo->head->is_target = 0;
1060                         ininfo = rinfo->targeti.in;
1061                         vino.ino = le64_to_cpu(ininfo->ino);
1062                         vino.snap = le64_to_cpu(ininfo->snapid);
1063                         in = ceph_get_inode(sb, vino);
1064                         err = fill_inode(in, &rinfo->targeti, NULL,
1065                                          session, req->r_request_started,
1066                                          req->r_fmode);
1067                         iput(in);
1068                 }
1069         }
1070 #endif
1071
1072         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1073                 dout("fill_trace reply is empty!\n");
1074                 if (rinfo->head->result == 0 && req->r_locked_dir)
1075                         ceph_invalidate_dir_request(req);
1076                 return 0;
1077         }
1078
1079         if (rinfo->head->is_dentry) {
1080                 struct inode *dir = req->r_locked_dir;
1081
1082                 if (dir) {
1083                         err = fill_inode(dir, NULL,
1084                                          &rinfo->diri, rinfo->dirfrag,
1085                                          session, req->r_request_started, -1,
1086                                          &req->r_caps_reservation);
1087                         if (err < 0)
1088                                 goto done;
1089                 } else {
1090                         WARN_ON_ONCE(1);
1091                 }
1092
1093                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1094                         struct qstr dname;
1095                         struct dentry *dn, *parent;
1096
1097                         BUG_ON(!rinfo->head->is_target);
1098                         BUG_ON(req->r_dentry);
1099
1100                         parent = d_find_any_alias(dir);
1101                         BUG_ON(!parent);
1102
1103                         dname.name = rinfo->dname;
1104                         dname.len = rinfo->dname_len;
1105                         dname.hash = full_name_hash(dname.name, dname.len);
1106                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1107                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1108 retry_lookup:
1109                         dn = d_lookup(parent, &dname);
1110                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1111                              parent, dname.len, dname.name, dn);
1112
1113                         if (!dn) {
1114                                 dn = d_alloc(parent, &dname);
1115                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1116                                      dname.len, dname.name, dn);
1117                                 if (dn == NULL) {
1118                                         dput(parent);
1119                                         err = -ENOMEM;
1120                                         goto done;
1121                                 }
1122                                 err = ceph_init_dentry(dn);
1123                                 if (err < 0) {
1124                                         dput(dn);
1125                                         dput(parent);
1126                                         goto done;
1127                                 }
1128                         } else if (d_really_is_positive(dn) &&
1129                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1130                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1131                                 dout(" dn %p points to wrong inode %p\n",
1132                                      dn, d_inode(dn));
1133                                 d_delete(dn);
1134                                 dput(dn);
1135                                 goto retry_lookup;
1136                         }
1137
1138                         req->r_dentry = dn;
1139                         dput(parent);
1140                 }
1141         }
1142
1143         if (rinfo->head->is_target) {
1144                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1145                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1146
1147                 in = ceph_get_inode(sb, vino);
1148                 if (IS_ERR(in)) {
1149                         err = PTR_ERR(in);
1150                         goto done;
1151                 }
1152                 req->r_target_inode = in;
1153
1154                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1155                                 session, req->r_request_started,
1156                                 (!req->r_aborted && rinfo->head->result == 0) ?
1157                                 req->r_fmode : -1,
1158                                 &req->r_caps_reservation);
1159                 if (err < 0) {
1160                         pr_err("fill_inode badness %p %llx.%llx\n",
1161                                 in, ceph_vinop(in));
1162                         goto done;
1163                 }
1164         }
1165
1166         /*
1167          * ignore null lease/binding on snapdir ENOENT, or else we
1168          * will have trouble splicing in the virtual snapdir later
1169          */
1170         if (rinfo->head->is_dentry && !req->r_aborted &&
1171             req->r_locked_dir &&
1172             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1173                                                fsc->mount_options->snapdir_name,
1174                                                req->r_dentry->d_name.len))) {
1175                 /*
1176                  * lookup link rename   : null -> possibly existing inode
1177                  * mknod symlink mkdir  : null -> new inode
1178                  * unlink               : linked -> null
1179                  */
1180                 struct inode *dir = req->r_locked_dir;
1181                 struct dentry *dn = req->r_dentry;
1182                 bool have_dir_cap, have_lease;
1183
1184                 BUG_ON(!dn);
1185                 BUG_ON(!dir);
1186                 BUG_ON(d_inode(dn->d_parent) != dir);
1187                 BUG_ON(ceph_ino(dir) !=
1188                        le64_to_cpu(rinfo->diri.in->ino));
1189                 BUG_ON(ceph_snap(dir) !=
1190                        le64_to_cpu(rinfo->diri.in->snapid));
1191
1192                 /* do we have a lease on the whole dir? */
1193                 have_dir_cap =
1194                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1195                          CEPH_CAP_FILE_SHARED);
1196
1197                 /* do we have a dn lease? */
1198                 have_lease = have_dir_cap ||
1199                         le32_to_cpu(rinfo->dlease->duration_ms);
1200                 if (!have_lease)
1201                         dout("fill_trace  no dentry lease or dir cap\n");
1202
1203                 /* rename? */
1204                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1205                         struct inode *olddir = req->r_old_dentry_dir;
1206                         BUG_ON(!olddir);
1207
1208                         dout(" src %p '%pd' dst %p '%pd'\n",
1209                              req->r_old_dentry,
1210                              req->r_old_dentry,
1211                              dn, dn);
1212                         dout("fill_trace doing d_move %p -> %p\n",
1213                              req->r_old_dentry, dn);
1214
1215                         d_move(req->r_old_dentry, dn);
1216                         dout(" src %p '%pd' dst %p '%pd'\n",
1217                              req->r_old_dentry,
1218                              req->r_old_dentry,
1219                              dn, dn);
1220
1221                         /* ensure target dentry is invalidated, despite
1222                            rehashing bug in vfs_rename_dir */
1223                         ceph_invalidate_dentry_lease(dn);
1224
1225                         /* d_move screws up sibling dentries' offsets */
1226                         ceph_dir_clear_ordered(dir);
1227                         ceph_dir_clear_ordered(olddir);
1228
1229                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1230                              ceph_dentry(req->r_old_dentry)->offset);
1231
1232                         dn = req->r_old_dentry;  /* use old_dentry */
1233                 }
1234
1235                 /* null dentry? */
1236                 if (!rinfo->head->is_target) {
1237                         dout("fill_trace null dentry\n");
1238                         if (d_really_is_positive(dn)) {
1239                                 ceph_dir_clear_ordered(dir);
1240                                 dout("d_delete %p\n", dn);
1241                                 d_delete(dn);
1242                         } else {
1243                                 dout("d_instantiate %p NULL\n", dn);
1244                                 d_instantiate(dn, NULL);
1245                                 if (have_lease && d_unhashed(dn))
1246                                         d_rehash(dn);
1247                                 update_dentry_lease(dn, rinfo->dlease,
1248                                                     session,
1249                                                     req->r_request_started);
1250                         }
1251                         goto done;
1252                 }
1253
1254                 /* attach proper inode */
1255                 if (d_really_is_negative(dn)) {
1256                         ceph_dir_clear_ordered(dir);
1257                         ihold(in);
1258                         dn = splice_dentry(dn, in, &have_lease);
1259                         if (IS_ERR(dn)) {
1260                                 err = PTR_ERR(dn);
1261                                 goto done;
1262                         }
1263                         req->r_dentry = dn;  /* may have spliced */
1264                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1265                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1266                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1267                              ceph_vinop(in));
1268                         have_lease = false;
1269                 }
1270
1271                 if (have_lease)
1272                         update_dentry_lease(dn, rinfo->dlease, session,
1273                                             req->r_request_started);
1274                 dout(" final dn %p\n", dn);
1275         } else if (!req->r_aborted &&
1276                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1277                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1278                 struct dentry *dn = req->r_dentry;
1279                 struct inode *dir = req->r_locked_dir;
1280
1281                 /* fill out a snapdir LOOKUPSNAP dentry */
1282                 BUG_ON(!dn);
1283                 BUG_ON(!dir);
1284                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1285                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1286                 ceph_dir_clear_ordered(dir);
1287                 ihold(in);
1288                 dn = splice_dentry(dn, in, NULL);
1289                 if (IS_ERR(dn)) {
1290                         err = PTR_ERR(dn);
1291                         goto done;
1292                 }
1293                 req->r_dentry = dn;  /* may have spliced */
1294         }
1295 done:
1296         dout("fill_trace done err=%d\n", err);
1297         return err;
1298 }
1299
1300 /*
1301  * Prepopulate our cache with readdir results, leases, etc.
1302  */
1303 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1304                                            struct ceph_mds_session *session)
1305 {
1306         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1307         int i, err = 0;
1308
1309         for (i = 0; i < rinfo->dir_nr; i++) {
1310                 struct ceph_vino vino;
1311                 struct inode *in;
1312                 int rc;
1313
1314                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1315                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1316
1317                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1318                 if (IS_ERR(in)) {
1319                         err = PTR_ERR(in);
1320                         dout("new_inode badness got %d\n", err);
1321                         continue;
1322                 }
1323                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1324                                 req->r_request_started, -1,
1325                                 &req->r_caps_reservation);
1326                 if (rc < 0) {
1327                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1328                         err = rc;
1329                         continue;
1330                 }
1331         }
1332
1333         return err;
1334 }
1335
1336 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1337                              struct ceph_mds_session *session)
1338 {
1339         struct dentry *parent = req->r_dentry;
1340         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1341         struct qstr dname;
1342         struct dentry *dn;
1343         struct inode *in;
1344         int err = 0, ret, i;
1345         struct inode *snapdir = NULL;
1346         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1347         struct ceph_dentry_info *di;
1348         u64 r_readdir_offset = req->r_readdir_offset;
1349         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1350
1351         if (rinfo->dir_dir &&
1352             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1353                 dout("readdir_prepopulate got new frag %x -> %x\n",
1354                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1355                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1356                 if (ceph_frag_is_leftmost(frag))
1357                         r_readdir_offset = 2;
1358                 else
1359                         r_readdir_offset = 0;
1360         }
1361
1362         if (req->r_aborted)
1363                 return readdir_prepopulate_inodes_only(req, session);
1364
1365         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1366                 snapdir = ceph_get_snapdir(d_inode(parent));
1367                 parent = d_find_alias(snapdir);
1368                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1369                      rinfo->dir_nr, parent);
1370         } else {
1371                 dout("readdir_prepopulate %d items under dn %p\n",
1372                      rinfo->dir_nr, parent);
1373                 if (rinfo->dir_dir)
1374                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1375         }
1376
1377         /* FIXME: release caps/leases if error occurs */
1378         for (i = 0; i < rinfo->dir_nr; i++) {
1379                 struct ceph_vino vino;
1380
1381                 dname.name = rinfo->dir_dname[i];
1382                 dname.len = rinfo->dir_dname_len[i];
1383                 dname.hash = full_name_hash(dname.name, dname.len);
1384
1385                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1386                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1387
1388 retry_lookup:
1389                 dn = d_lookup(parent, &dname);
1390                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1391                      parent, dname.len, dname.name, dn);
1392
1393                 if (!dn) {
1394                         dn = d_alloc(parent, &dname);
1395                         dout("d_alloc %p '%.*s' = %p\n", parent,
1396                              dname.len, dname.name, dn);
1397                         if (dn == NULL) {
1398                                 dout("d_alloc badness\n");
1399                                 err = -ENOMEM;
1400                                 goto out;
1401                         }
1402                         ret = ceph_init_dentry(dn);
1403                         if (ret < 0) {
1404                                 dput(dn);
1405                                 err = ret;
1406                                 goto out;
1407                         }
1408                 } else if (d_really_is_positive(dn) &&
1409                            (ceph_ino(d_inode(dn)) != vino.ino ||
1410                             ceph_snap(d_inode(dn)) != vino.snap)) {
1411                         dout(" dn %p points to wrong inode %p\n",
1412                              dn, d_inode(dn));
1413                         d_delete(dn);
1414                         dput(dn);
1415                         goto retry_lookup;
1416                 } else {
1417                         /* reorder parent's d_subdirs */
1418                         spin_lock(&parent->d_lock);
1419                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1420                         list_move(&dn->d_child, &parent->d_subdirs);
1421                         spin_unlock(&dn->d_lock);
1422                         spin_unlock(&parent->d_lock);
1423                 }
1424
1425                 /* inode */
1426                 if (d_really_is_positive(dn)) {
1427                         in = d_inode(dn);
1428                 } else {
1429                         in = ceph_get_inode(parent->d_sb, vino);
1430                         if (IS_ERR(in)) {
1431                                 dout("new_inode badness\n");
1432                                 d_drop(dn);
1433                                 dput(dn);
1434                                 err = PTR_ERR(in);
1435                                 goto out;
1436                         }
1437                 }
1438
1439                 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1440                                req->r_request_started, -1,
1441                                &req->r_caps_reservation) < 0) {
1442                         pr_err("fill_inode badness on %p\n", in);
1443                         if (d_really_is_negative(dn))
1444                                 iput(in);
1445                         d_drop(dn);
1446                         goto next_item;
1447                 }
1448
1449                 if (d_really_is_negative(dn)) {
1450                         struct dentry *realdn = splice_dentry(dn, in, NULL);
1451                         if (IS_ERR(realdn)) {
1452                                 err = PTR_ERR(realdn);
1453                                 d_drop(dn);
1454                                 dn = NULL;
1455                                 goto next_item;
1456                         }
1457                         dn = realdn;
1458                 }
1459
1460                 di = dn->d_fsdata;
1461                 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1462
1463                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1464                                     req->r_session,
1465                                     req->r_request_started);
1466 next_item:
1467                 if (dn)
1468                         dput(dn);
1469         }
1470         if (err == 0)
1471                 req->r_did_prepopulate = true;
1472
1473 out:
1474         if (snapdir) {
1475                 iput(snapdir);
1476                 dput(parent);
1477         }
1478         dout("readdir_prepopulate done\n");
1479         return err;
1480 }
1481
1482 int ceph_inode_set_size(struct inode *inode, loff_t size)
1483 {
1484         struct ceph_inode_info *ci = ceph_inode(inode);
1485         int ret = 0;
1486
1487         spin_lock(&ci->i_ceph_lock);
1488         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1489         inode->i_size = size;
1490         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1491
1492         /* tell the MDS if we are approaching max_size */
1493         if ((size << 1) >= ci->i_max_size &&
1494             (ci->i_reported_size << 1) < ci->i_max_size)
1495                 ret = 1;
1496
1497         spin_unlock(&ci->i_ceph_lock);
1498         return ret;
1499 }
1500
1501 /*
1502  * Write back inode data in a worker thread.  (This can't be done
1503  * in the message handler context.)
1504  */
1505 void ceph_queue_writeback(struct inode *inode)
1506 {
1507         ihold(inode);
1508         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1509                        &ceph_inode(inode)->i_wb_work)) {
1510                 dout("ceph_queue_writeback %p\n", inode);
1511         } else {
1512                 dout("ceph_queue_writeback %p failed\n", inode);
1513                 iput(inode);
1514         }
1515 }
1516
1517 static void ceph_writeback_work(struct work_struct *work)
1518 {
1519         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1520                                                   i_wb_work);
1521         struct inode *inode = &ci->vfs_inode;
1522
1523         dout("writeback %p\n", inode);
1524         filemap_fdatawrite(&inode->i_data);
1525         iput(inode);
1526 }
1527
1528 /*
1529  * queue an async invalidation
1530  */
1531 void ceph_queue_invalidate(struct inode *inode)
1532 {
1533         ihold(inode);
1534         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1535                        &ceph_inode(inode)->i_pg_inv_work)) {
1536                 dout("ceph_queue_invalidate %p\n", inode);
1537         } else {
1538                 dout("ceph_queue_invalidate %p failed\n", inode);
1539                 iput(inode);
1540         }
1541 }
1542
1543 /*
1544  * Invalidate inode pages in a worker thread.  (This can't be done
1545  * in the message handler context.)
1546  */
1547 static void ceph_invalidate_work(struct work_struct *work)
1548 {
1549         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1550                                                   i_pg_inv_work);
1551         struct inode *inode = &ci->vfs_inode;
1552         u32 orig_gen;
1553         int check = 0;
1554
1555         mutex_lock(&ci->i_truncate_mutex);
1556         spin_lock(&ci->i_ceph_lock);
1557         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1558              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1559         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1560                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1561                         check = 1;
1562                 spin_unlock(&ci->i_ceph_lock);
1563                 mutex_unlock(&ci->i_truncate_mutex);
1564                 goto out;
1565         }
1566         orig_gen = ci->i_rdcache_gen;
1567         spin_unlock(&ci->i_ceph_lock);
1568
1569         truncate_pagecache(inode, 0);
1570
1571         spin_lock(&ci->i_ceph_lock);
1572         if (orig_gen == ci->i_rdcache_gen &&
1573             orig_gen == ci->i_rdcache_revoking) {
1574                 dout("invalidate_pages %p gen %d successful\n", inode,
1575                      ci->i_rdcache_gen);
1576                 ci->i_rdcache_revoking--;
1577                 check = 1;
1578         } else {
1579                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1580                      inode, orig_gen, ci->i_rdcache_gen,
1581                      ci->i_rdcache_revoking);
1582                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1583                         check = 1;
1584         }
1585         spin_unlock(&ci->i_ceph_lock);
1586         mutex_unlock(&ci->i_truncate_mutex);
1587 out:
1588         if (check)
1589                 ceph_check_caps(ci, 0, NULL);
1590         iput(inode);
1591 }
1592
1593
1594 /*
1595  * called by trunc_wq;
1596  *
1597  * We also truncate in a separate thread as well.
1598  */
1599 static void ceph_vmtruncate_work(struct work_struct *work)
1600 {
1601         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1602                                                   i_vmtruncate_work);
1603         struct inode *inode = &ci->vfs_inode;
1604
1605         dout("vmtruncate_work %p\n", inode);
1606         __ceph_do_pending_vmtruncate(inode);
1607         iput(inode);
1608 }
1609
1610 /*
1611  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1612  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1613  */
1614 void ceph_queue_vmtruncate(struct inode *inode)
1615 {
1616         struct ceph_inode_info *ci = ceph_inode(inode);
1617
1618         ihold(inode);
1619
1620         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1621                        &ci->i_vmtruncate_work)) {
1622                 dout("ceph_queue_vmtruncate %p\n", inode);
1623         } else {
1624                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1625                      inode, ci->i_truncate_pending);
1626                 iput(inode);
1627         }
1628 }
1629
1630 /*
1631  * Make sure any pending truncation is applied before doing anything
1632  * that may depend on it.
1633  */
1634 void __ceph_do_pending_vmtruncate(struct inode *inode)
1635 {
1636         struct ceph_inode_info *ci = ceph_inode(inode);
1637         u64 to;
1638         int wrbuffer_refs, finish = 0;
1639
1640         mutex_lock(&ci->i_truncate_mutex);
1641 retry:
1642         spin_lock(&ci->i_ceph_lock);
1643         if (ci->i_truncate_pending == 0) {
1644                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1645                 spin_unlock(&ci->i_ceph_lock);
1646                 mutex_unlock(&ci->i_truncate_mutex);
1647                 return;
1648         }
1649
1650         /*
1651          * make sure any dirty snapped pages are flushed before we
1652          * possibly truncate them.. so write AND block!
1653          */
1654         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1655                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1656                      inode);
1657                 spin_unlock(&ci->i_ceph_lock);
1658                 filemap_write_and_wait_range(&inode->i_data, 0,
1659                                              inode->i_sb->s_maxbytes);
1660                 goto retry;
1661         }
1662
1663         /* there should be no reader or writer */
1664         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1665
1666         to = ci->i_truncate_size;
1667         wrbuffer_refs = ci->i_wrbuffer_ref;
1668         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1669              ci->i_truncate_pending, to);
1670         spin_unlock(&ci->i_ceph_lock);
1671
1672         truncate_pagecache(inode, to);
1673
1674         spin_lock(&ci->i_ceph_lock);
1675         if (to == ci->i_truncate_size) {
1676                 ci->i_truncate_pending = 0;
1677                 finish = 1;
1678         }
1679         spin_unlock(&ci->i_ceph_lock);
1680         if (!finish)
1681                 goto retry;
1682
1683         mutex_unlock(&ci->i_truncate_mutex);
1684
1685         if (wrbuffer_refs == 0)
1686                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1687
1688         wake_up_all(&ci->i_cap_wq);
1689 }
1690
1691 /*
1692  * symlinks
1693  */
1694 static const struct inode_operations ceph_symlink_iops = {
1695         .readlink = generic_readlink,
1696         .follow_link = simple_follow_link,
1697         .setattr = ceph_setattr,
1698         .getattr = ceph_getattr,
1699         .setxattr = ceph_setxattr,
1700         .getxattr = ceph_getxattr,
1701         .listxattr = ceph_listxattr,
1702         .removexattr = ceph_removexattr,
1703 };
1704
1705 /*
1706  * setattr
1707  */
1708 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1709 {
1710         struct inode *inode = d_inode(dentry);
1711         struct ceph_inode_info *ci = ceph_inode(inode);
1712         const unsigned int ia_valid = attr->ia_valid;
1713         struct ceph_mds_request *req;
1714         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1715         int issued;
1716         int release = 0, dirtied = 0;
1717         int mask = 0;
1718         int err = 0;
1719         int inode_dirty_flags = 0;
1720
1721         if (ceph_snap(inode) != CEPH_NOSNAP)
1722                 return -EROFS;
1723
1724         err = inode_change_ok(inode, attr);
1725         if (err != 0)
1726                 return err;
1727
1728         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1729                                        USE_AUTH_MDS);
1730         if (IS_ERR(req))
1731                 return PTR_ERR(req);
1732
1733         spin_lock(&ci->i_ceph_lock);
1734         issued = __ceph_caps_issued(ci, NULL);
1735         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1736
1737         if (ia_valid & ATTR_UID) {
1738                 dout("setattr %p uid %d -> %d\n", inode,
1739                      from_kuid(&init_user_ns, inode->i_uid),
1740                      from_kuid(&init_user_ns, attr->ia_uid));
1741                 if (issued & CEPH_CAP_AUTH_EXCL) {
1742                         inode->i_uid = attr->ia_uid;
1743                         dirtied |= CEPH_CAP_AUTH_EXCL;
1744                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1745                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1746                         req->r_args.setattr.uid = cpu_to_le32(
1747                                 from_kuid(&init_user_ns, attr->ia_uid));
1748                         mask |= CEPH_SETATTR_UID;
1749                         release |= CEPH_CAP_AUTH_SHARED;
1750                 }
1751         }
1752         if (ia_valid & ATTR_GID) {
1753                 dout("setattr %p gid %d -> %d\n", inode,
1754                      from_kgid(&init_user_ns, inode->i_gid),
1755                      from_kgid(&init_user_ns, attr->ia_gid));
1756                 if (issued & CEPH_CAP_AUTH_EXCL) {
1757                         inode->i_gid = attr->ia_gid;
1758                         dirtied |= CEPH_CAP_AUTH_EXCL;
1759                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1760                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1761                         req->r_args.setattr.gid = cpu_to_le32(
1762                                 from_kgid(&init_user_ns, attr->ia_gid));
1763                         mask |= CEPH_SETATTR_GID;
1764                         release |= CEPH_CAP_AUTH_SHARED;
1765                 }
1766         }
1767         if (ia_valid & ATTR_MODE) {
1768                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1769                      attr->ia_mode);
1770                 if (issued & CEPH_CAP_AUTH_EXCL) {
1771                         inode->i_mode = attr->ia_mode;
1772                         dirtied |= CEPH_CAP_AUTH_EXCL;
1773                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1774                            attr->ia_mode != inode->i_mode) {
1775                         inode->i_mode = attr->ia_mode;
1776                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1777                         mask |= CEPH_SETATTR_MODE;
1778                         release |= CEPH_CAP_AUTH_SHARED;
1779                 }
1780         }
1781
1782         if (ia_valid & ATTR_ATIME) {
1783                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1784                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1785                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1786                 if (issued & CEPH_CAP_FILE_EXCL) {
1787                         ci->i_time_warp_seq++;
1788                         inode->i_atime = attr->ia_atime;
1789                         dirtied |= CEPH_CAP_FILE_EXCL;
1790                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1791                            timespec_compare(&inode->i_atime,
1792                                             &attr->ia_atime) < 0) {
1793                         inode->i_atime = attr->ia_atime;
1794                         dirtied |= CEPH_CAP_FILE_WR;
1795                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1796                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1797                         ceph_encode_timespec(&req->r_args.setattr.atime,
1798                                              &attr->ia_atime);
1799                         mask |= CEPH_SETATTR_ATIME;
1800                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1801                                 CEPH_CAP_FILE_WR;
1802                 }
1803         }
1804         if (ia_valid & ATTR_MTIME) {
1805                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1806                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1807                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1808                 if (issued & CEPH_CAP_FILE_EXCL) {
1809                         ci->i_time_warp_seq++;
1810                         inode->i_mtime = attr->ia_mtime;
1811                         dirtied |= CEPH_CAP_FILE_EXCL;
1812                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1813                            timespec_compare(&inode->i_mtime,
1814                                             &attr->ia_mtime) < 0) {
1815                         inode->i_mtime = attr->ia_mtime;
1816                         dirtied |= CEPH_CAP_FILE_WR;
1817                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1818                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1819                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1820                                              &attr->ia_mtime);
1821                         mask |= CEPH_SETATTR_MTIME;
1822                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1823                                 CEPH_CAP_FILE_WR;
1824                 }
1825         }
1826         if (ia_valid & ATTR_SIZE) {
1827                 dout("setattr %p size %lld -> %lld\n", inode,
1828                      inode->i_size, attr->ia_size);
1829                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1830                     attr->ia_size > inode->i_size) {
1831                         inode->i_size = attr->ia_size;
1832                         inode->i_blocks =
1833                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1834                         inode->i_ctime = attr->ia_ctime;
1835                         ci->i_reported_size = attr->ia_size;
1836                         dirtied |= CEPH_CAP_FILE_EXCL;
1837                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1838                            attr->ia_size != inode->i_size) {
1839                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1840                         req->r_args.setattr.old_size =
1841                                 cpu_to_le64(inode->i_size);
1842                         mask |= CEPH_SETATTR_SIZE;
1843                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1844                                 CEPH_CAP_FILE_WR;
1845                 }
1846         }
1847
1848         /* these do nothing */
1849         if (ia_valid & ATTR_CTIME) {
1850                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1851                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1852                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1853                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1854                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1855                      only ? "ctime only" : "ignored");
1856                 inode->i_ctime = attr->ia_ctime;
1857                 if (only) {
1858                         /*
1859                          * if kernel wants to dirty ctime but nothing else,
1860                          * we need to choose a cap to dirty under, or do
1861                          * a almost-no-op setattr
1862                          */
1863                         if (issued & CEPH_CAP_AUTH_EXCL)
1864                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1865                         else if (issued & CEPH_CAP_FILE_EXCL)
1866                                 dirtied |= CEPH_CAP_FILE_EXCL;
1867                         else if (issued & CEPH_CAP_XATTR_EXCL)
1868                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1869                         else
1870                                 mask |= CEPH_SETATTR_CTIME;
1871                 }
1872         }
1873         if (ia_valid & ATTR_FILE)
1874                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1875
1876         if (dirtied) {
1877                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1878                 inode->i_ctime = CURRENT_TIME;
1879         }
1880
1881         release &= issued;
1882         spin_unlock(&ci->i_ceph_lock);
1883
1884         if (inode_dirty_flags)
1885                 __mark_inode_dirty(inode, inode_dirty_flags);
1886
1887         if (ia_valid & ATTR_MODE) {
1888                 err = posix_acl_chmod(inode, attr->ia_mode);
1889                 if (err)
1890                         goto out_put;
1891         }
1892
1893         if (mask) {
1894                 req->r_inode = inode;
1895                 ihold(inode);
1896                 req->r_inode_drop = release;
1897                 req->r_args.setattr.mask = cpu_to_le32(mask);
1898                 req->r_num_caps = 1;
1899                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1900         }
1901         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1902              ceph_cap_string(dirtied), mask);
1903
1904         ceph_mdsc_put_request(req);
1905         if (mask & CEPH_SETATTR_SIZE)
1906                 __ceph_do_pending_vmtruncate(inode);
1907         return err;
1908 out_put:
1909         ceph_mdsc_put_request(req);
1910         return err;
1911 }
1912
1913 /*
1914  * Verify that we have a lease on the given mask.  If not,
1915  * do a getattr against an mds.
1916  */
1917 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1918                       int mask, bool force)
1919 {
1920         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1921         struct ceph_mds_client *mdsc = fsc->mdsc;
1922         struct ceph_mds_request *req;
1923         int err;
1924
1925         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1926                 dout("do_getattr inode %p SNAPDIR\n", inode);
1927                 return 0;
1928         }
1929
1930         dout("do_getattr inode %p mask %s mode 0%o\n",
1931              inode, ceph_cap_string(mask), inode->i_mode);
1932         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1933                 return 0;
1934
1935         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1936         if (IS_ERR(req))
1937                 return PTR_ERR(req);
1938         req->r_inode = inode;
1939         ihold(inode);
1940         req->r_num_caps = 1;
1941         req->r_args.getattr.mask = cpu_to_le32(mask);
1942         req->r_locked_page = locked_page;
1943         err = ceph_mdsc_do_request(mdsc, NULL, req);
1944         if (locked_page && err == 0) {
1945                 u64 inline_version = req->r_reply_info.targeti.inline_version;
1946                 if (inline_version == 0) {
1947                         /* the reply is supposed to contain inline data */
1948                         err = -EINVAL;
1949                 } else if (inline_version == CEPH_INLINE_NONE) {
1950                         err = -ENODATA;
1951                 } else {
1952                         err = req->r_reply_info.targeti.inline_len;
1953                 }
1954         }
1955         ceph_mdsc_put_request(req);
1956         dout("do_getattr result=%d\n", err);
1957         return err;
1958 }
1959
1960
1961 /*
1962  * Check inode permissions.  We verify we have a valid value for
1963  * the AUTH cap, then call the generic handler.
1964  */
1965 int ceph_permission(struct inode *inode, int mask)
1966 {
1967         int err;
1968
1969         if (mask & MAY_NOT_BLOCK)
1970                 return -ECHILD;
1971
1972         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1973
1974         if (!err)
1975                 err = generic_permission(inode, mask);
1976         return err;
1977 }
1978
1979 /*
1980  * Get all attributes.  Hopefully somedata we'll have a statlite()
1981  * and can limit the fields we require to be accurate.
1982  */
1983 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1984                  struct kstat *stat)
1985 {
1986         struct inode *inode = d_inode(dentry);
1987         struct ceph_inode_info *ci = ceph_inode(inode);
1988         int err;
1989
1990         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
1991         if (!err) {
1992                 generic_fillattr(inode, stat);
1993                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1994                 if (ceph_snap(inode) != CEPH_NOSNAP)
1995                         stat->dev = ceph_snap(inode);
1996                 else
1997                         stat->dev = 0;
1998                 if (S_ISDIR(inode->i_mode)) {
1999                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2000                                                 RBYTES))
2001                                 stat->size = ci->i_rbytes;
2002                         else
2003                                 stat->size = ci->i_files + ci->i_subdirs;
2004                         stat->blocks = 0;
2005                         stat->blksize = 65536;
2006                 }
2007         }
2008         return err;
2009 }