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         atomic64_set(&ci->i_ordered_count, 1);
 393         atomic64_set(&ci->i_release_count, 1);
 394         atomic64_set(&ci->i_complete_seq[0], 0);
 395         atomic64_set(&ci->i_complete_seq[1], 0);
 396         ci->i_symlink = NULL;
 397
 398         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 399
 400         ci->i_fragtree = RB_ROOT;
 401         mutex_init(&ci->i_fragtree_mutex);
 402
 403         ci->i_xattrs.blob = NULL;
 404         ci->i_xattrs.prealloc_blob = NULL;
 405         ci->i_xattrs.dirty = false;
 406         ci->i_xattrs.index = RB_ROOT;
 407         ci->i_xattrs.count = 0;
 408         ci->i_xattrs.names_size = 0;
 409         ci->i_xattrs.vals_size = 0;
 410         ci->i_xattrs.version = 0;
 411         ci->i_xattrs.index_version = 0;
 412
 413         ci->i_caps = RB_ROOT;
 414         ci->i_auth_cap = NULL;
 415         ci->i_dirty_caps = 0;
 416         ci->i_flushing_caps = 0;
 417         INIT_LIST_HEAD(&ci->i_dirty_item);
 418         INIT_LIST_HEAD(&ci->i_flushing_item);
 419         ci->i_prealloc_cap_flush = NULL;
 420         ci->i_cap_flush_tree = RB_ROOT;
 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                 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
 756                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 757                 ci->i_layout = info->layout;
 758
 759                 queue_trunc = ceph_fill_file_size(inode, issued,
 760                                         le32_to_cpu(info->truncate_seq),
 761                                         le64_to_cpu(info->truncate_size),
 762                                         le64_to_cpu(info->size));
 763                 /* only update max_size on auth cap */
 764                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 765                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 766                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 767                                         le64_to_cpu(info->max_size));
 768                         ci->i_max_size = le64_to_cpu(info->max_size);
 769                 }
 770         }
 771
 772         /* xattrs */
 773         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 774         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 775             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 776                 if (ci->i_xattrs.blob)
 777                         ceph_buffer_put(ci->i_xattrs.blob);
 778                 ci->i_xattrs.blob = xattr_blob;
 779                 if (xattr_blob)
 780                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 781                                iinfo->xattr_data, iinfo->xattr_len);
 782                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 783                 ceph_forget_all_cached_acls(inode);
 784                 xattr_blob = NULL;
 785         }
 786
 787         inode->i_mapping->a_ops = &ceph_aops;
 788
 789         switch (inode->i_mode & S_IFMT) {
 790         case S_IFIFO:
 791         case S_IFBLK:
 792         case S_IFCHR:
 793         case S_IFSOCK:
 794                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 795                 inode->i_op = &ceph_file_iops;
 796                 break;
 797         case S_IFREG:
 798                 inode->i_op = &ceph_file_iops;
 799                 inode->i_fop = &ceph_file_fops;
 800                 break;
 801         case S_IFLNK:
 802                 inode->i_op = &ceph_symlink_iops;
 803                 if (!ci->i_symlink) {
 804                         u32 symlen = iinfo->symlink_len;
 805                         char *sym;
 806
 807                         spin_unlock(&ci->i_ceph_lock);
 808
 809                         err = -EINVAL;
 810                         if (WARN_ON(symlen != inode->i_size))
 811                                 goto out;
 812
 813                         err = -ENOMEM;
 814                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 815                         if (!sym)
 816                                 goto out;
 817
 818                         spin_lock(&ci->i_ceph_lock);
 819                         if (!ci->i_symlink)
 820                                 ci->i_symlink = sym;
 821                         else
 822                                 kfree(sym); /* lost a race */
 823                 }
 824                 inode->i_link = ci->i_symlink;
 825                 break;
 826         case S_IFDIR:
 827                 inode->i_op = &ceph_dir_iops;
 828                 inode->i_fop = &ceph_dir_fops;
 829
 830                 ci->i_dir_layout = iinfo->dir_layout;
 831
 832                 ci->i_files = le64_to_cpu(info->files);
 833                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 834                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 835                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 836                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 837                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 838                 break;
 839         default:
 840                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 841                        ceph_vinop(inode), inode->i_mode);
 842         }
 843
 844         /* were we issued a capability? */
 845         if (info->cap.caps) {
 846                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 847                         unsigned caps = le32_to_cpu(info->cap.caps);
 848                         ceph_add_cap(inode, session,
 849                                      le64_to_cpu(info->cap.cap_id),
 850                                      cap_fmode, caps,
 851                                      le32_to_cpu(info->cap.wanted),
 852                                      le32_to_cpu(info->cap.seq),
 853                                      le32_to_cpu(info->cap.mseq),
 854                                      le64_to_cpu(info->cap.realm),
 855                                      info->cap.flags, &new_cap);
 856
 857                         /* set dir completion flag? */
 858                         if (S_ISDIR(inode->i_mode) &&
 859                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 860                             (caps & CEPH_CAP_FILE_SHARED) &&
 861                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 862                             !__ceph_dir_is_complete(ci)) {
 863                                 dout(" marking %p complete (empty)\n", inode);
 864                                 i_size_write(inode, 0);
 865                                 __ceph_dir_set_complete(ci,
 866                                         atomic64_read(&ci->i_release_count),
 867                                         atomic64_read(&ci->i_ordered_count));
 868                         }
 869
 870                         wake = true;
 871                 } else {
 872                         dout(" %p got snap_caps %s\n", inode,
 873                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 874                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 875                         if (cap_fmode >= 0)
 876                                 __ceph_get_fmode(ci, cap_fmode);
 877                 }
 878         } else if (cap_fmode >= 0) {
 879                 pr_warn("mds issued no caps on %llx.%llx\n",
 880                            ceph_vinop(inode));
 881                 __ceph_get_fmode(ci, cap_fmode);
 882         }
 883
 884         if (iinfo->inline_version > 0 &&
 885             iinfo->inline_version >= ci->i_inline_version) {
 886                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 887                 ci->i_inline_version = iinfo->inline_version;
 888                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 889                     (locked_page ||
 890                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 891                         fill_inline = true;
 892         }
 893
 894         spin_unlock(&ci->i_ceph_lock);
 895
 896         if (fill_inline)
 897                 ceph_fill_inline_data(inode, locked_page,
 898                                       iinfo->inline_data, iinfo->inline_len);
 899
 900         if (wake)
 901                 wake_up_all(&ci->i_cap_wq);
 902
 903         /* queue truncate if we saw i_size decrease */
 904         if (queue_trunc)
 905                 ceph_queue_vmtruncate(inode);
 906
 907         /* populate frag tree */
 908         if (S_ISDIR(inode->i_mode))
 909                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 910
 911         /* update delegation info? */
 912         if (dirinfo)
 913                 ceph_fill_dirfrag(inode, dirinfo);
 914
 915         err = 0;
 916 out:
 917         if (new_cap)
 918                 ceph_put_cap(mdsc, new_cap);
 919         if (xattr_blob)
 920                 ceph_buffer_put(xattr_blob);
 921         return err;
 922 }
 923
 924 /*
 925  * caller should hold session s_mutex.
 926  */
 927 static void update_dentry_lease(struct dentry *dentry,
 928                                 struct ceph_mds_reply_lease *lease,
 929                                 struct ceph_mds_session *session,
 930                                 unsigned long from_time)
 931 {
 932         struct ceph_dentry_info *di = ceph_dentry(dentry);
 933         long unsigned duration = le32_to_cpu(lease->duration_ms);
 934         long unsigned ttl = from_time + (duration * HZ) / 1000;
 935         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 936         struct inode *dir;
 937
 938         /* only track leases on regular dentries */
 939         if (dentry->d_op != &ceph_dentry_ops)
 940                 return;
 941
 942         spin_lock(&dentry->d_lock);
 943         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 944              dentry, duration, ttl);
 945
 946         /* make lease_rdcache_gen match directory */
 947         dir = d_inode(dentry->d_parent);
 948         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 949
 950         if (duration == 0)
 951                 goto out_unlock;
 952
 953         if (di->lease_gen == session->s_cap_gen &&
 954             time_before(ttl, dentry->d_time))
 955                 goto out_unlock;  /* we already have a newer lease. */
 956
 957         if (di->lease_session && di->lease_session != session)
 958                 goto out_unlock;
 959
 960         ceph_dentry_lru_touch(dentry);
 961
 962         if (!di->lease_session)
 963                 di->lease_session = ceph_get_mds_session(session);
 964         di->lease_gen = session->s_cap_gen;
 965         di->lease_seq = le32_to_cpu(lease->seq);
 966         di->lease_renew_after = half_ttl;
 967         di->lease_renew_from = 0;
 968         dentry->d_time = ttl;
 969 out_unlock:
 970         spin_unlock(&dentry->d_lock);
 971         return;
 972 }
 973
 974 /*
 975  * splice a dentry to an inode.
 976  * caller must hold directory i_mutex for this to be safe.
 977  *
 978  * we will only rehash the resulting dentry if @prehash is
 979  * true; @prehash will be set to false (for the benefit of
 980  * the caller) if we fail.
 981  */
 982 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 983                                     bool *prehash)
 984 {
 985         struct dentry *realdn;
 986
 987         BUG_ON(d_inode(dn));
 988
 989         /* dn must be unhashed */
 990         if (!d_unhashed(dn))
 991                 d_drop(dn);
 992         realdn = d_splice_alias(in, dn);
 993         if (IS_ERR(realdn)) {
 994                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 995                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 996                 if (prehash)
 997                         *prehash = false; /* don't rehash on error */
 998                 dn = realdn; /* note realdn contains the error */
 999                 goto out;
1000         } else if (realdn) {
1001                 dout("dn %p (%d) spliced with %p (%d) "
1002                      "inode %p ino %llx.%llx\n",
1003                      dn, d_count(dn),
1004                      realdn, d_count(realdn),
1005                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1006                 dput(dn);
1007                 dn = realdn;
1008         } else {
1009                 BUG_ON(!ceph_dentry(dn));
1010                 dout("dn %p attached to %p ino %llx.%llx\n",
1011                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1012         }
1013         if ((!prehash || *prehash) && d_unhashed(dn))
1014                 d_rehash(dn);
1015 out:
1016         return dn;
1017 }
1018
1019 /*
1020  * Incorporate results into the local cache.  This is either just
1021  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1022  * after a lookup).
1023  *
1024  * A reply may contain
1025  *         a directory inode along with a dentry.
1026  *  and/or a target inode
1027  *
1028  * Called with snap_rwsem (read).
1029  */
1030 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1031                     struct ceph_mds_session *session)
1032 {
1033         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1034         struct inode *in = NULL;
1035         struct ceph_vino vino;
1036         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1037         int err = 0;
1038
1039         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1040              rinfo->head->is_dentry, rinfo->head->is_target);
1041
1042 #if 0
1043         /*
1044          * Debugging hook:
1045          *
1046          * If we resend completed ops to a recovering mds, we get no
1047          * trace.  Since that is very rare, pretend this is the case
1048          * to ensure the 'no trace' handlers in the callers behave.
1049          *
1050          * Fill in inodes unconditionally to avoid breaking cap
1051          * invariants.
1052          */
1053         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1054                 pr_info("fill_trace faking empty trace on %lld %s\n",
1055                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1056                 if (rinfo->head->is_dentry) {
1057                         rinfo->head->is_dentry = 0;
1058                         err = fill_inode(req->r_locked_dir,
1059                                          &rinfo->diri, rinfo->dirfrag,
1060                                          session, req->r_request_started, -1);
1061                 }
1062                 if (rinfo->head->is_target) {
1063                         rinfo->head->is_target = 0;
1064                         ininfo = rinfo->targeti.in;
1065                         vino.ino = le64_to_cpu(ininfo->ino);
1066                         vino.snap = le64_to_cpu(ininfo->snapid);
1067                         in = ceph_get_inode(sb, vino);
1068                         err = fill_inode(in, &rinfo->targeti, NULL,
1069                                          session, req->r_request_started,
1070                                          req->r_fmode);
1071                         iput(in);
1072                 }
1073         }
1074 #endif
1075
1076         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1077                 dout("fill_trace reply is empty!\n");
1078                 if (rinfo->head->result == 0 && req->r_locked_dir)
1079                         ceph_invalidate_dir_request(req);
1080                 return 0;
1081         }
1082
1083         if (rinfo->head->is_dentry) {
1084                 struct inode *dir = req->r_locked_dir;
1085
1086                 if (dir) {
1087                         err = fill_inode(dir, NULL,
1088                                          &rinfo->diri, rinfo->dirfrag,
1089                                          session, req->r_request_started, -1,
1090                                          &req->r_caps_reservation);
1091                         if (err < 0)
1092                                 goto done;
1093                 } else {
1094                         WARN_ON_ONCE(1);
1095                 }
1096
1097                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1098                         struct qstr dname;
1099                         struct dentry *dn, *parent;
1100
1101                         BUG_ON(!rinfo->head->is_target);
1102                         BUG_ON(req->r_dentry);
1103
1104                         parent = d_find_any_alias(dir);
1105                         BUG_ON(!parent);
1106
1107                         dname.name = rinfo->dname;
1108                         dname.len = rinfo->dname_len;
1109                         dname.hash = full_name_hash(dname.name, dname.len);
1110                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1111                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1112 retry_lookup:
1113                         dn = d_lookup(parent, &dname);
1114                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1115                              parent, dname.len, dname.name, dn);
1116
1117                         if (!dn) {
1118                                 dn = d_alloc(parent, &dname);
1119                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1120                                      dname.len, dname.name, dn);
1121                                 if (dn == NULL) {
1122                                         dput(parent);
1123                                         err = -ENOMEM;
1124                                         goto done;
1125                                 }
1126                                 err = ceph_init_dentry(dn);
1127                                 if (err < 0) {
1128                                         dput(dn);
1129                                         dput(parent);
1130                                         goto done;
1131                                 }
1132                         } else if (d_really_is_positive(dn) &&
1133                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1134                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1135                                 dout(" dn %p points to wrong inode %p\n",
1136                                      dn, d_inode(dn));
1137                                 d_delete(dn);
1138                                 dput(dn);
1139                                 goto retry_lookup;
1140                         }
1141
1142                         req->r_dentry = dn;
1143                         dput(parent);
1144                 }
1145         }
1146
1147         if (rinfo->head->is_target) {
1148                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1149                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1150
1151                 in = ceph_get_inode(sb, vino);
1152                 if (IS_ERR(in)) {
1153                         err = PTR_ERR(in);
1154                         goto done;
1155                 }
1156                 req->r_target_inode = in;
1157
1158                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1159                                 session, req->r_request_started,
1160                                 (!req->r_aborted && rinfo->head->result == 0) ?
1161                                 req->r_fmode : -1,
1162                                 &req->r_caps_reservation);
1163                 if (err < 0) {
1164                         pr_err("fill_inode badness %p %llx.%llx\n",
1165                                 in, ceph_vinop(in));
1166                         goto done;
1167                 }
1168         }
1169
1170         /*
1171          * ignore null lease/binding on snapdir ENOENT, or else we
1172          * will have trouble splicing in the virtual snapdir later
1173          */
1174         if (rinfo->head->is_dentry && !req->r_aborted &&
1175             req->r_locked_dir &&
1176             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1177                                                fsc->mount_options->snapdir_name,
1178                                                req->r_dentry->d_name.len))) {
1179                 /*
1180                  * lookup link rename   : null -> possibly existing inode
1181                  * mknod symlink mkdir  : null -> new inode
1182                  * unlink               : linked -> null
1183                  */
1184                 struct inode *dir = req->r_locked_dir;
1185                 struct dentry *dn = req->r_dentry;
1186                 bool have_dir_cap, have_lease;
1187
1188                 BUG_ON(!dn);
1189                 BUG_ON(!dir);
1190                 BUG_ON(d_inode(dn->d_parent) != dir);
1191                 BUG_ON(ceph_ino(dir) !=
1192                        le64_to_cpu(rinfo->diri.in->ino));
1193                 BUG_ON(ceph_snap(dir) !=
1194                        le64_to_cpu(rinfo->diri.in->snapid));
1195
1196                 /* do we have a lease on the whole dir? */
1197                 have_dir_cap =
1198                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1199                          CEPH_CAP_FILE_SHARED);
1200
1201                 /* do we have a dn lease? */
1202                 have_lease = have_dir_cap ||
1203                         le32_to_cpu(rinfo->dlease->duration_ms);
1204                 if (!have_lease)
1205                         dout("fill_trace  no dentry lease or dir cap\n");
1206
1207                 /* rename? */
1208                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1209                         struct inode *olddir = req->r_old_dentry_dir;
1210                         BUG_ON(!olddir);
1211
1212                         dout(" src %p '%pd' dst %p '%pd'\n",
1213                              req->r_old_dentry,
1214                              req->r_old_dentry,
1215                              dn, dn);
1216                         dout("fill_trace doing d_move %p -> %p\n",
1217                              req->r_old_dentry, dn);
1218
1219                         /* d_move screws up sibling dentries' offsets */
1220                         ceph_dir_clear_ordered(dir);
1221                         ceph_dir_clear_ordered(olddir);
1222
1223                         d_move(req->r_old_dentry, dn);
1224                         dout(" src %p '%pd' dst %p '%pd'\n",
1225                              req->r_old_dentry,
1226                              req->r_old_dentry,
1227                              dn, dn);
1228
1229                         /* ensure target dentry is invalidated, despite
1230                            rehashing bug in vfs_rename_dir */
1231                         ceph_invalidate_dentry_lease(dn);
1232
1233                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1234                              ceph_dentry(req->r_old_dentry)->offset);
1235
1236                         dn = req->r_old_dentry;  /* use old_dentry */
1237                 }
1238
1239                 /* null dentry? */
1240                 if (!rinfo->head->is_target) {
1241                         dout("fill_trace null dentry\n");
1242                         if (d_really_is_positive(dn)) {
1243                                 ceph_dir_clear_ordered(dir);
1244                                 dout("d_delete %p\n", dn);
1245                                 d_delete(dn);
1246                         } else {
1247                                 dout("d_instantiate %p NULL\n", dn);
1248                                 d_instantiate(dn, NULL);
1249                                 if (have_lease && d_unhashed(dn))
1250                                         d_rehash(dn);
1251                                 update_dentry_lease(dn, rinfo->dlease,
1252                                                     session,
1253                                                     req->r_request_started);
1254                         }
1255                         goto done;
1256                 }
1257
1258                 /* attach proper inode */
1259                 if (d_really_is_negative(dn)) {
1260                         ceph_dir_clear_ordered(dir);
1261                         ihold(in);
1262                         dn = splice_dentry(dn, in, &have_lease);
1263                         if (IS_ERR(dn)) {
1264                                 err = PTR_ERR(dn);
1265                                 goto done;
1266                         }
1267                         req->r_dentry = dn;  /* may have spliced */
1268                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1269                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1270                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1271                              ceph_vinop(in));
1272                         have_lease = false;
1273                 }
1274
1275                 if (have_lease)
1276                         update_dentry_lease(dn, rinfo->dlease, session,
1277                                             req->r_request_started);
1278                 dout(" final dn %p\n", dn);
1279         } else if (!req->r_aborted &&
1280                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1281                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1282                 struct dentry *dn = req->r_dentry;
1283                 struct inode *dir = req->r_locked_dir;
1284
1285                 /* fill out a snapdir LOOKUPSNAP dentry */
1286                 BUG_ON(!dn);
1287                 BUG_ON(!dir);
1288                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1289                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1290                 ceph_dir_clear_ordered(dir);
1291                 ihold(in);
1292                 dn = splice_dentry(dn, in, NULL);
1293                 if (IS_ERR(dn)) {
1294                         err = PTR_ERR(dn);
1295                         goto done;
1296                 }
1297                 req->r_dentry = dn;  /* may have spliced */
1298         }
1299 done:
1300         dout("fill_trace done err=%d\n", err);
1301         return err;
1302 }
1303
1304 /*
1305  * Prepopulate our cache with readdir results, leases, etc.
1306  */
1307 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1308                                            struct ceph_mds_session *session)
1309 {
1310         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1311         int i, err = 0;
1312
1313         for (i = 0; i < rinfo->dir_nr; i++) {
1314                 struct ceph_vino vino;
1315                 struct inode *in;
1316                 int rc;
1317
1318                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1319                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1320
1321                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1322                 if (IS_ERR(in)) {
1323                         err = PTR_ERR(in);
1324                         dout("new_inode badness got %d\n", err);
1325                         continue;
1326                 }
1327                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1328                                 req->r_request_started, -1,
1329                                 &req->r_caps_reservation);
1330                 if (rc < 0) {
1331                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1332                         err = rc;
1333                         continue;
1334                 }
1335         }
1336
1337         return err;
1338 }
1339
1340 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1341 {
1342         if (ctl->page) {
1343                 kunmap(ctl->page);
1344                 page_cache_release(ctl->page);
1345                 ctl->page = NULL;
1346         }
1347 }
1348
1349 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1350                               struct ceph_readdir_cache_control *ctl,
1351                               struct ceph_mds_request *req)
1352 {
1353         struct ceph_inode_info *ci = ceph_inode(dir);
1354         unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry*);
1355         unsigned idx = ctl->index % nsize;
1356         pgoff_t pgoff = ctl->index / nsize;
1357
1358         if (!ctl->page || pgoff != page_index(ctl->page)) {
1359                 ceph_readdir_cache_release(ctl);
1360                 ctl->page  = grab_cache_page(&dir->i_data, pgoff);
1361                 if (!ctl->page) {
1362                         ctl->index = -1;
1363                         return -ENOMEM;
1364                 }
1365                 /* reading/filling the cache are serialized by
1366                  * i_mutex, no need to use page lock */
1367                 unlock_page(ctl->page);
1368                 ctl->dentries = kmap(ctl->page);
1369         }
1370
1371         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1372             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1373                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1374                 ctl->dentries[idx] = dn;
1375                 ctl->index++;
1376         } else {
1377                 dout("disable readdir cache\n");
1378                 ctl->index = -1;
1379         }
1380         return 0;
1381 }
1382
1383 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1384                              struct ceph_mds_session *session)
1385 {
1386         struct dentry *parent = req->r_dentry;
1387         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1388         struct qstr dname;
1389         struct dentry *dn;
1390         struct inode *in;
1391         int err = 0, ret, i;
1392         struct inode *snapdir = NULL;
1393         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1394         struct ceph_dentry_info *di;
1395         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1396         struct ceph_readdir_cache_control cache_ctl = {};
1397
1398         if (req->r_aborted)
1399                 return readdir_prepopulate_inodes_only(req, session);
1400
1401         if (rinfo->dir_dir &&
1402             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1403                 dout("readdir_prepopulate got new frag %x -> %x\n",
1404                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1405                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1406                 if (ceph_frag_is_leftmost(frag))
1407                         req->r_readdir_offset = 2;
1408                 else
1409                         req->r_readdir_offset = 0;
1410         }
1411
1412         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1413                 snapdir = ceph_get_snapdir(d_inode(parent));
1414                 parent = d_find_alias(snapdir);
1415                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1416                      rinfo->dir_nr, parent);
1417         } else {
1418                 dout("readdir_prepopulate %d items under dn %p\n",
1419                      rinfo->dir_nr, parent);
1420                 if (rinfo->dir_dir)
1421                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1422         }
1423
1424         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1425                 /* note dir version at start of readdir so we can tell
1426                  * if any dentries get dropped */
1427                 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1428                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1429                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1430                 req->r_readdir_cache_idx = 0;
1431         }
1432
1433         cache_ctl.index = req->r_readdir_cache_idx;
1434
1435         /* FIXME: release caps/leases if error occurs */
1436         for (i = 0; i < rinfo->dir_nr; i++) {
1437                 struct ceph_vino vino;
1438
1439                 dname.name = rinfo->dir_dname[i];
1440                 dname.len = rinfo->dir_dname_len[i];
1441                 dname.hash = full_name_hash(dname.name, dname.len);
1442
1443                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1444                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1445
1446 retry_lookup:
1447                 dn = d_lookup(parent, &dname);
1448                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1449                      parent, dname.len, dname.name, dn);
1450
1451                 if (!dn) {
1452                         dn = d_alloc(parent, &dname);
1453                         dout("d_alloc %p '%.*s' = %p\n", parent,
1454                              dname.len, dname.name, dn);
1455                         if (dn == NULL) {
1456                                 dout("d_alloc badness\n");
1457                                 err = -ENOMEM;
1458                                 goto out;
1459                         }
1460                         ret = ceph_init_dentry(dn);
1461                         if (ret < 0) {
1462                                 dput(dn);
1463                                 err = ret;
1464                                 goto out;
1465                         }
1466                 } else if (d_really_is_positive(dn) &&
1467                            (ceph_ino(d_inode(dn)) != vino.ino ||
1468                             ceph_snap(d_inode(dn)) != vino.snap)) {
1469                         dout(" dn %p points to wrong inode %p\n",
1470                              dn, d_inode(dn));
1471                         d_delete(dn);
1472                         dput(dn);
1473                         goto retry_lookup;
1474                 }
1475
1476                 /* inode */
1477                 if (d_really_is_positive(dn)) {
1478                         in = d_inode(dn);
1479                 } else {
1480                         in = ceph_get_inode(parent->d_sb, vino);
1481                         if (IS_ERR(in)) {
1482                                 dout("new_inode badness\n");
1483                                 d_drop(dn);
1484                                 dput(dn);
1485                                 err = PTR_ERR(in);
1486                                 goto out;
1487                         }
1488                 }
1489
1490                 ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1491                                  req->r_request_started, -1,
1492                                  &req->r_caps_reservation);
1493                 if (ret < 0) {
1494                         pr_err("fill_inode badness on %p\n", in);
1495                         if (d_really_is_negative(dn))
1496                                 iput(in);
1497                         d_drop(dn);
1498                         err = ret;
1499                         goto next_item;
1500                 }
1501
1502                 if (d_really_is_negative(dn)) {
1503                         struct dentry *realdn = splice_dentry(dn, in, NULL);
1504                         if (IS_ERR(realdn)) {
1505                                 err = PTR_ERR(realdn);
1506                                 d_drop(dn);
1507                                 dn = NULL;
1508                                 goto next_item;
1509                         }
1510                         dn = realdn;
1511                 }
1512
1513                 di = dn->d_fsdata;
1514                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1515
1516                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1517                                     req->r_session,
1518                                     req->r_request_started);
1519
1520                 if (err == 0 && cache_ctl.index >= 0) {
1521                         ret = fill_readdir_cache(d_inode(parent), dn,
1522                                                  &cache_ctl, req);
1523                         if (ret < 0)
1524                                 err = ret;
1525                 }
1526 next_item:
1527                 if (dn)
1528                         dput(dn);
1529         }
1530 out:
1531         if (err == 0) {
1532                 req->r_did_prepopulate = true;
1533                 req->r_readdir_cache_idx = cache_ctl.index;
1534         }
1535         ceph_readdir_cache_release(&cache_ctl);
1536         if (snapdir) {
1537                 iput(snapdir);
1538                 dput(parent);
1539         }
1540         dout("readdir_prepopulate done\n");
1541         return err;
1542 }
1543
1544 int ceph_inode_set_size(struct inode *inode, loff_t size)
1545 {
1546         struct ceph_inode_info *ci = ceph_inode(inode);
1547         int ret = 0;
1548
1549         spin_lock(&ci->i_ceph_lock);
1550         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1551         inode->i_size = size;
1552         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1553
1554         /* tell the MDS if we are approaching max_size */
1555         if ((size << 1) >= ci->i_max_size &&
1556             (ci->i_reported_size << 1) < ci->i_max_size)
1557                 ret = 1;
1558
1559         spin_unlock(&ci->i_ceph_lock);
1560         return ret;
1561 }
1562
1563 /*
1564  * Write back inode data in a worker thread.  (This can't be done
1565  * in the message handler context.)
1566  */
1567 void ceph_queue_writeback(struct inode *inode)
1568 {
1569         ihold(inode);
1570         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1571                        &ceph_inode(inode)->i_wb_work)) {
1572                 dout("ceph_queue_writeback %p\n", inode);
1573         } else {
1574                 dout("ceph_queue_writeback %p failed\n", inode);
1575                 iput(inode);
1576         }
1577 }
1578
1579 static void ceph_writeback_work(struct work_struct *work)
1580 {
1581         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1582                                                   i_wb_work);
1583         struct inode *inode = &ci->vfs_inode;
1584
1585         dout("writeback %p\n", inode);
1586         filemap_fdatawrite(&inode->i_data);
1587         iput(inode);
1588 }
1589
1590 /*
1591  * queue an async invalidation
1592  */
1593 void ceph_queue_invalidate(struct inode *inode)
1594 {
1595         ihold(inode);
1596         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1597                        &ceph_inode(inode)->i_pg_inv_work)) {
1598                 dout("ceph_queue_invalidate %p\n", inode);
1599         } else {
1600                 dout("ceph_queue_invalidate %p failed\n", inode);
1601                 iput(inode);
1602         }
1603 }
1604
1605 /*
1606  * Invalidate inode pages in a worker thread.  (This can't be done
1607  * in the message handler context.)
1608  */
1609 static void ceph_invalidate_work(struct work_struct *work)
1610 {
1611         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1612                                                   i_pg_inv_work);
1613         struct inode *inode = &ci->vfs_inode;
1614         u32 orig_gen;
1615         int check = 0;
1616
1617         mutex_lock(&ci->i_truncate_mutex);
1618         spin_lock(&ci->i_ceph_lock);
1619         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1620              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1621         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1622                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1623                         check = 1;
1624                 spin_unlock(&ci->i_ceph_lock);
1625                 mutex_unlock(&ci->i_truncate_mutex);
1626                 goto out;
1627         }
1628         orig_gen = ci->i_rdcache_gen;
1629         spin_unlock(&ci->i_ceph_lock);
1630
1631         truncate_pagecache(inode, 0);
1632
1633         spin_lock(&ci->i_ceph_lock);
1634         if (orig_gen == ci->i_rdcache_gen &&
1635             orig_gen == ci->i_rdcache_revoking) {
1636                 dout("invalidate_pages %p gen %d successful\n", inode,
1637                      ci->i_rdcache_gen);
1638                 ci->i_rdcache_revoking--;
1639                 check = 1;
1640         } else {
1641                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1642                      inode, orig_gen, ci->i_rdcache_gen,
1643                      ci->i_rdcache_revoking);
1644                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1645                         check = 1;
1646         }
1647         spin_unlock(&ci->i_ceph_lock);
1648         mutex_unlock(&ci->i_truncate_mutex);
1649 out:
1650         if (check)
1651                 ceph_check_caps(ci, 0, NULL);
1652         iput(inode);
1653 }
1654
1655
1656 /*
1657  * called by trunc_wq;
1658  *
1659  * We also truncate in a separate thread as well.
1660  */
1661 static void ceph_vmtruncate_work(struct work_struct *work)
1662 {
1663         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1664                                                   i_vmtruncate_work);
1665         struct inode *inode = &ci->vfs_inode;
1666
1667         dout("vmtruncate_work %p\n", inode);
1668         __ceph_do_pending_vmtruncate(inode);
1669         iput(inode);
1670 }
1671
1672 /*
1673  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1674  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1675  */
1676 void ceph_queue_vmtruncate(struct inode *inode)
1677 {
1678         struct ceph_inode_info *ci = ceph_inode(inode);
1679
1680         ihold(inode);
1681
1682         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1683                        &ci->i_vmtruncate_work)) {
1684                 dout("ceph_queue_vmtruncate %p\n", inode);
1685         } else {
1686                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1687                      inode, ci->i_truncate_pending);
1688                 iput(inode);
1689         }
1690 }
1691
1692 /*
1693  * Make sure any pending truncation is applied before doing anything
1694  * that may depend on it.
1695  */
1696 void __ceph_do_pending_vmtruncate(struct inode *inode)
1697 {
1698         struct ceph_inode_info *ci = ceph_inode(inode);
1699         u64 to;
1700         int wrbuffer_refs, finish = 0;
1701
1702         mutex_lock(&ci->i_truncate_mutex);
1703 retry:
1704         spin_lock(&ci->i_ceph_lock);
1705         if (ci->i_truncate_pending == 0) {
1706                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1707                 spin_unlock(&ci->i_ceph_lock);
1708                 mutex_unlock(&ci->i_truncate_mutex);
1709                 return;
1710         }
1711
1712         /*
1713          * make sure any dirty snapped pages are flushed before we
1714          * possibly truncate them.. so write AND block!
1715          */
1716         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1717                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1718                      inode);
1719                 spin_unlock(&ci->i_ceph_lock);
1720                 filemap_write_and_wait_range(&inode->i_data, 0,
1721                                              inode->i_sb->s_maxbytes);
1722                 goto retry;
1723         }
1724
1725         /* there should be no reader or writer */
1726         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1727
1728         to = ci->i_truncate_size;
1729         wrbuffer_refs = ci->i_wrbuffer_ref;
1730         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1731              ci->i_truncate_pending, to);
1732         spin_unlock(&ci->i_ceph_lock);
1733
1734         truncate_pagecache(inode, to);
1735
1736         spin_lock(&ci->i_ceph_lock);
1737         if (to == ci->i_truncate_size) {
1738                 ci->i_truncate_pending = 0;
1739                 finish = 1;
1740         }
1741         spin_unlock(&ci->i_ceph_lock);
1742         if (!finish)
1743                 goto retry;
1744
1745         mutex_unlock(&ci->i_truncate_mutex);
1746
1747         if (wrbuffer_refs == 0)
1748                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1749
1750         wake_up_all(&ci->i_cap_wq);
1751 }
1752
1753 /*
1754  * symlinks
1755  */
1756 static const struct inode_operations ceph_symlink_iops = {
1757         .readlink = generic_readlink,
1758         .follow_link = simple_follow_link,
1759         .setattr = ceph_setattr,
1760         .getattr = ceph_getattr,
1761         .setxattr = ceph_setxattr,
1762         .getxattr = ceph_getxattr,
1763         .listxattr = ceph_listxattr,
1764         .removexattr = ceph_removexattr,
1765 };
1766
1767 /*
1768  * setattr
1769  */
1770 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1771 {
1772         struct inode *inode = d_inode(dentry);
1773         struct ceph_inode_info *ci = ceph_inode(inode);
1774         const unsigned int ia_valid = attr->ia_valid;
1775         struct ceph_mds_request *req;
1776         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1777         struct ceph_cap_flush *prealloc_cf;
1778         int issued;
1779         int release = 0, dirtied = 0;
1780         int mask = 0;
1781         int err = 0;
1782         int inode_dirty_flags = 0;
1783         bool lock_snap_rwsem = false;
1784
1785         if (ceph_snap(inode) != CEPH_NOSNAP)
1786                 return -EROFS;
1787
1788         err = inode_change_ok(inode, attr);
1789         if (err != 0)
1790                 return err;
1791
1792         prealloc_cf = ceph_alloc_cap_flush();
1793         if (!prealloc_cf)
1794                 return -ENOMEM;
1795
1796         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1797                                        USE_AUTH_MDS);
1798         if (IS_ERR(req)) {
1799                 ceph_free_cap_flush(prealloc_cf);
1800                 return PTR_ERR(req);
1801         }
1802
1803         spin_lock(&ci->i_ceph_lock);
1804         issued = __ceph_caps_issued(ci, NULL);
1805
1806         if (!ci->i_head_snapc &&
1807             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1808                 lock_snap_rwsem = true;
1809                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1810                         spin_unlock(&ci->i_ceph_lock);
1811                         down_read(&mdsc->snap_rwsem);
1812                         spin_lock(&ci->i_ceph_lock);
1813                         issued = __ceph_caps_issued(ci, NULL);
1814                 }
1815         }
1816
1817         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1818
1819         if (ia_valid & ATTR_UID) {
1820                 dout("setattr %p uid %d -> %d\n", inode,
1821                      from_kuid(&init_user_ns, inode->i_uid),
1822                      from_kuid(&init_user_ns, attr->ia_uid));
1823                 if (issued & CEPH_CAP_AUTH_EXCL) {
1824                         inode->i_uid = attr->ia_uid;
1825                         dirtied |= CEPH_CAP_AUTH_EXCL;
1826                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1827                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1828                         req->r_args.setattr.uid = cpu_to_le32(
1829                                 from_kuid(&init_user_ns, attr->ia_uid));
1830                         mask |= CEPH_SETATTR_UID;
1831                         release |= CEPH_CAP_AUTH_SHARED;
1832                 }
1833         }
1834         if (ia_valid & ATTR_GID) {
1835                 dout("setattr %p gid %d -> %d\n", inode,
1836                      from_kgid(&init_user_ns, inode->i_gid),
1837                      from_kgid(&init_user_ns, attr->ia_gid));
1838                 if (issued & CEPH_CAP_AUTH_EXCL) {
1839                         inode->i_gid = attr->ia_gid;
1840                         dirtied |= CEPH_CAP_AUTH_EXCL;
1841                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1842                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1843                         req->r_args.setattr.gid = cpu_to_le32(
1844                                 from_kgid(&init_user_ns, attr->ia_gid));
1845                         mask |= CEPH_SETATTR_GID;
1846                         release |= CEPH_CAP_AUTH_SHARED;
1847                 }
1848         }
1849         if (ia_valid & ATTR_MODE) {
1850                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1851                      attr->ia_mode);
1852                 if (issued & CEPH_CAP_AUTH_EXCL) {
1853                         inode->i_mode = attr->ia_mode;
1854                         dirtied |= CEPH_CAP_AUTH_EXCL;
1855                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1856                            attr->ia_mode != inode->i_mode) {
1857                         inode->i_mode = attr->ia_mode;
1858                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1859                         mask |= CEPH_SETATTR_MODE;
1860                         release |= CEPH_CAP_AUTH_SHARED;
1861                 }
1862         }
1863
1864         if (ia_valid & ATTR_ATIME) {
1865                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1866                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1867                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1868                 if (issued & CEPH_CAP_FILE_EXCL) {
1869                         ci->i_time_warp_seq++;
1870                         inode->i_atime = attr->ia_atime;
1871                         dirtied |= CEPH_CAP_FILE_EXCL;
1872                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1873                            timespec_compare(&inode->i_atime,
1874                                             &attr->ia_atime) < 0) {
1875                         inode->i_atime = attr->ia_atime;
1876                         dirtied |= CEPH_CAP_FILE_WR;
1877                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1878                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1879                         ceph_encode_timespec(&req->r_args.setattr.atime,
1880                                              &attr->ia_atime);
1881                         mask |= CEPH_SETATTR_ATIME;
1882                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1883                                 CEPH_CAP_FILE_WR;
1884                 }
1885         }
1886         if (ia_valid & ATTR_MTIME) {
1887                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1888                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1889                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1890                 if (issued & CEPH_CAP_FILE_EXCL) {
1891                         ci->i_time_warp_seq++;
1892                         inode->i_mtime = attr->ia_mtime;
1893                         dirtied |= CEPH_CAP_FILE_EXCL;
1894                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1895                            timespec_compare(&inode->i_mtime,
1896                                             &attr->ia_mtime) < 0) {
1897                         inode->i_mtime = attr->ia_mtime;
1898                         dirtied |= CEPH_CAP_FILE_WR;
1899                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1900                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1901                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1902                                              &attr->ia_mtime);
1903                         mask |= CEPH_SETATTR_MTIME;
1904                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1905                                 CEPH_CAP_FILE_WR;
1906                 }
1907         }
1908         if (ia_valid & ATTR_SIZE) {
1909                 dout("setattr %p size %lld -> %lld\n", inode,
1910                      inode->i_size, attr->ia_size);
1911                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1912                     attr->ia_size > inode->i_size) {
1913                         inode->i_size = attr->ia_size;
1914                         inode->i_blocks =
1915                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1916                         inode->i_ctime = attr->ia_ctime;
1917                         ci->i_reported_size = attr->ia_size;
1918                         dirtied |= CEPH_CAP_FILE_EXCL;
1919                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1920                            attr->ia_size != inode->i_size) {
1921                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1922                         req->r_args.setattr.old_size =
1923                                 cpu_to_le64(inode->i_size);
1924                         mask |= CEPH_SETATTR_SIZE;
1925                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1926                                 CEPH_CAP_FILE_WR;
1927                 }
1928         }
1929
1930         /* these do nothing */
1931         if (ia_valid & ATTR_CTIME) {
1932                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1933                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1934                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1935                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1936                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1937                      only ? "ctime only" : "ignored");
1938                 inode->i_ctime = attr->ia_ctime;
1939                 if (only) {
1940                         /*
1941                          * if kernel wants to dirty ctime but nothing else,
1942                          * we need to choose a cap to dirty under, or do
1943                          * a almost-no-op setattr
1944                          */
1945                         if (issued & CEPH_CAP_AUTH_EXCL)
1946                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1947                         else if (issued & CEPH_CAP_FILE_EXCL)
1948                                 dirtied |= CEPH_CAP_FILE_EXCL;
1949                         else if (issued & CEPH_CAP_XATTR_EXCL)
1950                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1951                         else
1952                                 mask |= CEPH_SETATTR_CTIME;
1953                 }
1954         }
1955         if (ia_valid & ATTR_FILE)
1956                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1957
1958         if (dirtied) {
1959                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
1960                                                            &prealloc_cf);
1961                 inode->i_ctime = CURRENT_TIME;
1962         }
1963
1964         release &= issued;
1965         spin_unlock(&ci->i_ceph_lock);
1966         if (lock_snap_rwsem)
1967                 up_read(&mdsc->snap_rwsem);
1968
1969         if (inode_dirty_flags)
1970                 __mark_inode_dirty(inode, inode_dirty_flags);
1971
1972         if (ia_valid & ATTR_MODE) {
1973                 err = posix_acl_chmod(inode, attr->ia_mode);
1974                 if (err)
1975                         goto out_put;
1976         }
1977
1978         if (mask) {
1979                 req->r_inode = inode;
1980                 ihold(inode);
1981                 req->r_inode_drop = release;
1982                 req->r_args.setattr.mask = cpu_to_le32(mask);
1983                 req->r_num_caps = 1;
1984                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1985         }
1986         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1987              ceph_cap_string(dirtied), mask);
1988
1989         ceph_mdsc_put_request(req);
1990         if (mask & CEPH_SETATTR_SIZE)
1991                 __ceph_do_pending_vmtruncate(inode);
1992         ceph_free_cap_flush(prealloc_cf);
1993         return err;
1994 out_put:
1995         ceph_mdsc_put_request(req);
1996         ceph_free_cap_flush(prealloc_cf);
1997         return err;
1998 }
1999
2000 /*
2001  * Verify that we have a lease on the given mask.  If not,
2002  * do a getattr against an mds.
2003  */
2004 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2005                       int mask, bool force)
2006 {
2007         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2008         struct ceph_mds_client *mdsc = fsc->mdsc;
2009         struct ceph_mds_request *req;
2010         int err;
2011
2012         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2013                 dout("do_getattr inode %p SNAPDIR\n", inode);
2014                 return 0;
2015         }
2016
2017         dout("do_getattr inode %p mask %s mode 0%o\n",
2018              inode, ceph_cap_string(mask), inode->i_mode);
2019         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2020                 return 0;
2021
2022         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2023         if (IS_ERR(req))
2024                 return PTR_ERR(req);
2025         req->r_inode = inode;
2026         ihold(inode);
2027         req->r_num_caps = 1;
2028         req->r_args.getattr.mask = cpu_to_le32(mask);
2029         req->r_locked_page = locked_page;
2030         err = ceph_mdsc_do_request(mdsc, NULL, req);
2031         if (locked_page && err == 0) {
2032                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2033                 if (inline_version == 0) {
2034                         /* the reply is supposed to contain inline data */
2035                         err = -EINVAL;
2036                 } else if (inline_version == CEPH_INLINE_NONE) {
2037                         err = -ENODATA;
2038                 } else {
2039                         err = req->r_reply_info.targeti.inline_len;
2040                 }
2041         }
2042         ceph_mdsc_put_request(req);
2043         dout("do_getattr result=%d\n", err);
2044         return err;
2045 }
2046
2047
2048 /*
2049  * Check inode permissions.  We verify we have a valid value for
2050  * the AUTH cap, then call the generic handler.
2051  */
2052 int ceph_permission(struct inode *inode, int mask)
2053 {
2054         int err;
2055
2056         if (mask & MAY_NOT_BLOCK)
2057                 return -ECHILD;
2058
2059         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2060
2061         if (!err)
2062                 err = generic_permission(inode, mask);
2063         return err;
2064 }
2065
2066 /*
2067  * Get all attributes.  Hopefully somedata we'll have a statlite()
2068  * and can limit the fields we require to be accurate.
2069  */
2070 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2071                  struct kstat *stat)
2072 {
2073         struct inode *inode = d_inode(dentry);
2074         struct ceph_inode_info *ci = ceph_inode(inode);
2075         int err;
2076
2077         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2078         if (!err) {
2079                 generic_fillattr(inode, stat);
2080                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2081                 if (ceph_snap(inode) != CEPH_NOSNAP)
2082                         stat->dev = ceph_snap(inode);
2083                 else
2084                         stat->dev = 0;
2085                 if (S_ISDIR(inode->i_mode)) {
2086                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2087                                                 RBYTES))
2088                                 stat->size = ci->i_rbytes;
2089                         else
2090                                 stat->size = ci->i_files + ci->i_subdirs;
2091                         stat->blocks = 0;
2092                         stat->blksize = 65536;
2093                 }
2094         }
2095         return err;
2096 }