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/namei.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/posix_acl.h>
  13 #include <linux/random.h>
  14
  15 #include "super.h"
  16 #include "mds_client.h"
  17 #include "cache.h"
  18 #include <linux/ceph/decode.h>
  19
  20 /*
  21  * Ceph inode operations
  22  *
  23  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  24  * setattr, etc.), xattr helpers, and helpers for assimilating
  25  * metadata returned by the MDS into our cache.
  26  *
  27  * Also define helpers for doing asynchronous writeback, invalidation,
  28  * and truncation for the benefit of those who can't afford to block
  29  * (typically because they are in the message handler path).
  30  */
  31
  32 static const struct inode_operations ceph_symlink_iops;
  33
  34 static void ceph_invalidate_work(struct work_struct *work);
  35 static void ceph_writeback_work(struct work_struct *work);
  36 static void ceph_vmtruncate_work(struct work_struct *work);
  37
  38 /*
  39  * find or create an inode, given the ceph ino number
  40  */
  41 static int ceph_set_ino_cb(struct inode *inode, void *data)
  42 {
  43         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  44         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  45         return 0;
  46 }
  47
  48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  49 {
  50         struct inode *inode;
  51         ino_t t = ceph_vino_to_ino(vino);
  52
  53         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  54         if (inode == NULL)
  55                 return ERR_PTR(-ENOMEM);
  56         if (inode->i_state & I_NEW) {
  57                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  58                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  59                 unlock_new_inode(inode);
  60         }
  61
  62         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  63              vino.snap, inode);
  64         return inode;
  65 }
  66
  67 /*
  68  * get/constuct snapdir inode for a given directory
  69  */
  70 struct inode *ceph_get_snapdir(struct inode *parent)
  71 {
  72         struct ceph_vino vino = {
  73                 .ino = ceph_ino(parent),
  74                 .snap = CEPH_SNAPDIR,
  75         };
  76         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  77         struct ceph_inode_info *ci = ceph_inode(inode);
  78
  79         BUG_ON(!S_ISDIR(parent->i_mode));
  80         if (IS_ERR(inode))
  81                 return inode;
  82         inode->i_mode = parent->i_mode;
  83         inode->i_uid = parent->i_uid;
  84         inode->i_gid = parent->i_gid;
  85         inode->i_op = &ceph_dir_iops;
  86         inode->i_fop = &ceph_dir_fops;
  87         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  88         ci->i_rbytes = 0;
  89         return inode;
  90 }
  91
  92 const struct inode_operations ceph_file_iops = {
  93         .permission = ceph_permission,
  94         .setattr = ceph_setattr,
  95         .getattr = ceph_getattr,
  96         .setxattr = ceph_setxattr,
  97         .getxattr = ceph_getxattr,
  98         .listxattr = ceph_listxattr,
  99         .removexattr = ceph_removexattr,
 100         .get_acl = ceph_get_acl,
 101         .set_acl = ceph_set_acl,
 102 };
 103
 104
 105 /*
 106  * We use a 'frag tree' to keep track of the MDS's directory fragments
 107  * for a given inode (usually there is just a single fragment).  We
 108  * need to know when a child frag is delegated to a new MDS, or when
 109  * it is flagged as replicated, so we can direct our requests
 110  * accordingly.
 111  */
 112
 113 /*
 114  * find/create a frag in the tree
 115  */
 116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 117                                                     u32 f)
 118 {
 119         struct rb_node **p;
 120         struct rb_node *parent = NULL;
 121         struct ceph_inode_frag *frag;
 122         int c;
 123
 124         p = &ci->i_fragtree.rb_node;
 125         while (*p) {
 126                 parent = *p;
 127                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 128                 c = ceph_frag_compare(f, frag->frag);
 129                 if (c < 0)
 130                         p = &(*p)->rb_left;
 131                 else if (c > 0)
 132                         p = &(*p)->rb_right;
 133                 else
 134                         return frag;
 135         }
 136
 137         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 138         if (!frag) {
 139                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 140                        "frag %x\n", &ci->vfs_inode,
 141                        ceph_vinop(&ci->vfs_inode), f);
 142                 return ERR_PTR(-ENOMEM);
 143         }
 144         frag->frag = f;
 145         frag->split_by = 0;
 146         frag->mds = -1;
 147         frag->ndist = 0;
 148
 149         rb_link_node(&frag->node, parent, p);
 150         rb_insert_color(&frag->node, &ci->i_fragtree);
 151
 152         dout("get_or_create_frag added %llx.%llx frag %x\n",
 153              ceph_vinop(&ci->vfs_inode), f);
 154         return frag;
 155 }
 156
 157 /*
 158  * find a specific frag @f
 159  */
 160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 161 {
 162         struct rb_node *n = ci->i_fragtree.rb_node;
 163
 164         while (n) {
 165                 struct ceph_inode_frag *frag =
 166                         rb_entry(n, struct ceph_inode_frag, node);
 167                 int c = ceph_frag_compare(f, frag->frag);
 168                 if (c < 0)
 169                         n = n->rb_left;
 170                 else if (c > 0)
 171                         n = n->rb_right;
 172                 else
 173                         return frag;
 174         }
 175         return NULL;
 176 }
 177
 178 /*
 179  * Choose frag containing the given value @v.  If @pfrag is
 180  * specified, copy the frag delegation info to the caller if
 181  * it is present.
 182  */
 183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 184                               struct ceph_inode_frag *pfrag, int *found)
 185 {
 186         u32 t = ceph_frag_make(0, 0);
 187         struct ceph_inode_frag *frag;
 188         unsigned nway, i;
 189         u32 n;
 190
 191         if (found)
 192                 *found = 0;
 193
 194         while (1) {
 195                 WARN_ON(!ceph_frag_contains_value(t, v));
 196                 frag = __ceph_find_frag(ci, t);
 197                 if (!frag)
 198                         break; /* t is a leaf */
 199                 if (frag->split_by == 0) {
 200                         if (pfrag)
 201                                 memcpy(pfrag, frag, sizeof(*pfrag));
 202                         if (found)
 203                                 *found = 1;
 204                         break;
 205                 }
 206
 207                 /* choose child */
 208                 nway = 1 << frag->split_by;
 209                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 210                      frag->split_by, nway);
 211                 for (i = 0; i < nway; i++) {
 212                         n = ceph_frag_make_child(t, frag->split_by, i);
 213                         if (ceph_frag_contains_value(n, v)) {
 214                                 t = n;
 215                                 break;
 216                         }
 217                 }
 218                 BUG_ON(i == nway);
 219         }
 220         dout("choose_frag(%x) = %x\n", v, t);
 221
 222         return t;
 223 }
 224
 225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 226                      struct ceph_inode_frag *pfrag, int *found)
 227 {
 228         u32 ret;
 229         mutex_lock(&ci->i_fragtree_mutex);
 230         ret = __ceph_choose_frag(ci, v, pfrag, found);
 231         mutex_unlock(&ci->i_fragtree_mutex);
 232         return ret;
 233 }
 234
 235 /*
 236  * Process dirfrag (delegation) info from the mds.  Include leaf
 237  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 238  * branches/splits are included in i_fragtree)
 239  */
 240 static int ceph_fill_dirfrag(struct inode *inode,
 241                              struct ceph_mds_reply_dirfrag *dirinfo)
 242 {
 243         struct ceph_inode_info *ci = ceph_inode(inode);
 244         struct ceph_inode_frag *frag;
 245         u32 id = le32_to_cpu(dirinfo->frag);
 246         int mds = le32_to_cpu(dirinfo->auth);
 247         int ndist = le32_to_cpu(dirinfo->ndist);
 248         int diri_auth = -1;
 249         int i;
 250         int err = 0;
 251
 252         spin_lock(&ci->i_ceph_lock);
 253         if (ci->i_auth_cap)
 254                 diri_auth = ci->i_auth_cap->mds;
 255         spin_unlock(&ci->i_ceph_lock);
 256
 257         mutex_lock(&ci->i_fragtree_mutex);
 258         if (ndist == 0 && mds == diri_auth) {
 259                 /* no delegation info needed. */
 260                 frag = __ceph_find_frag(ci, id);
 261                 if (!frag)
 262                         goto out;
 263                 if (frag->split_by == 0) {
 264                         /* tree leaf, remove */
 265                         dout("fill_dirfrag removed %llx.%llx frag %x"
 266                              " (no ref)\n", ceph_vinop(inode), id);
 267                         rb_erase(&frag->node, &ci->i_fragtree);
 268                         kfree(frag);
 269                 } else {
 270                         /* tree branch, keep and clear */
 271                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 272                              " referral\n", ceph_vinop(inode), id);
 273                         frag->mds = -1;
 274                         frag->ndist = 0;
 275                 }
 276                 goto out;
 277         }
 278
 279
 280         /* find/add this frag to store mds delegation info */
 281         frag = __get_or_create_frag(ci, id);
 282         if (IS_ERR(frag)) {
 283                 /* this is not the end of the world; we can continue
 284                    with bad/inaccurate delegation info */
 285                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 286                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 287                 err = -ENOMEM;
 288                 goto out;
 289         }
 290
 291         frag->mds = mds;
 292         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 293         for (i = 0; i < frag->ndist; i++)
 294                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 295         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 296              ceph_vinop(inode), frag->frag, frag->ndist);
 297
 298 out:
 299         mutex_unlock(&ci->i_fragtree_mutex);
 300         return err;
 301 }
 302
 303 static int ceph_fill_fragtree(struct inode *inode,
 304                               struct ceph_frag_tree_head *fragtree,
 305                               struct ceph_mds_reply_dirfrag *dirinfo)
 306 {
 307         struct ceph_inode_info *ci = ceph_inode(inode);
 308         struct ceph_inode_frag *frag;
 309         struct rb_node *rb_node;
 310         int i;
 311         u32 id, nsplits;
 312         bool update = false;
 313
 314         mutex_lock(&ci->i_fragtree_mutex);
 315         nsplits = le32_to_cpu(fragtree->nsplits);
 316         if (nsplits) {
 317                 i = prandom_u32() % nsplits;
 318                 id = le32_to_cpu(fragtree->splits[i].frag);
 319                 if (!__ceph_find_frag(ci, id))
 320                         update = true;
 321         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 322                 rb_node = rb_first(&ci->i_fragtree);
 323                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 324                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 325                         update = true;
 326         }
 327         if (!update && dirinfo) {
 328                 id = le32_to_cpu(dirinfo->frag);
 329                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 330                         update = true;
 331         }
 332         if (!update)
 333                 goto out_unlock;
 334
 335         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 336         rb_node = rb_first(&ci->i_fragtree);
 337         for (i = 0; i < nsplits; i++) {
 338                 id = le32_to_cpu(fragtree->splits[i].frag);
 339                 frag = NULL;
 340                 while (rb_node) {
 341                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 342                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 343                                 if (frag->frag != id)
 344                                         frag = NULL;
 345                                 else
 346                                         rb_node = rb_next(rb_node);
 347                                 break;
 348                         }
 349                         rb_node = rb_next(rb_node);
 350                         rb_erase(&frag->node, &ci->i_fragtree);
 351                         kfree(frag);
 352                         frag = NULL;
 353                 }
 354                 if (!frag) {
 355                         frag = __get_or_create_frag(ci, id);
 356                         if (IS_ERR(frag))
 357                                 continue;
 358                 }
 359                 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
 360                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 361         }
 362         while (rb_node) {
 363                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 364                 rb_node = rb_next(rb_node);
 365                 rb_erase(&frag->node, &ci->i_fragtree);
 366                 kfree(frag);
 367         }
 368 out_unlock:
 369         mutex_unlock(&ci->i_fragtree_mutex);
 370         return 0;
 371 }
 372
 373 /*
 374  * initialize a newly allocated inode.
 375  */
 376 struct inode *ceph_alloc_inode(struct super_block *sb)
 377 {
 378         struct ceph_inode_info *ci;
 379         int i;
 380
 381         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 382         if (!ci)
 383                 return NULL;
 384
 385         dout("alloc_inode %p\n", &ci->vfs_inode);
 386
 387         spin_lock_init(&ci->i_ceph_lock);
 388
 389         ci->i_version = 0;
 390         ci->i_inline_version = 0;
 391         ci->i_time_warp_seq = 0;
 392         ci->i_ceph_flags = 0;
 393         ci->i_ordered_count = 0;
 394         atomic_set(&ci->i_release_count, 1);
 395         atomic_set(&ci->i_complete_count, 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_cap_flush_seq = 0;
 420         ci->i_cap_flush_last_tid = 0;
 421         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 422         init_waitqueue_head(&ci->i_cap_wq);
 423         ci->i_hold_caps_min = 0;
 424         ci->i_hold_caps_max = 0;
 425         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 426         INIT_LIST_HEAD(&ci->i_cap_snaps);
 427         ci->i_head_snapc = NULL;
 428         ci->i_snap_caps = 0;
 429
 430         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 431                 ci->i_nr_by_mode[i] = 0;
 432
 433         mutex_init(&ci->i_truncate_mutex);
 434         ci->i_truncate_seq = 0;
 435         ci->i_truncate_size = 0;
 436         ci->i_truncate_pending = 0;
 437
 438         ci->i_max_size = 0;
 439         ci->i_reported_size = 0;
 440         ci->i_wanted_max_size = 0;
 441         ci->i_requested_max_size = 0;
 442
 443         ci->i_pin_ref = 0;
 444         ci->i_rd_ref = 0;
 445         ci->i_rdcache_ref = 0;
 446         ci->i_wr_ref = 0;
 447         ci->i_wb_ref = 0;
 448         ci->i_wrbuffer_ref = 0;
 449         ci->i_wrbuffer_ref_head = 0;
 450         ci->i_shared_gen = 0;
 451         ci->i_rdcache_gen = 0;
 452         ci->i_rdcache_revoking = 0;
 453
 454         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 455         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 456         spin_lock_init(&ci->i_unsafe_lock);
 457
 458         ci->i_snap_realm = NULL;
 459         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 460         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 461
 462         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 463         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 464
 465         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 466
 467         ceph_fscache_inode_init(ci);
 468
 469         return &ci->vfs_inode;
 470 }
 471
 472 static void ceph_i_callback(struct rcu_head *head)
 473 {
 474         struct inode *inode = container_of(head, struct inode, i_rcu);
 475         struct ceph_inode_info *ci = ceph_inode(inode);
 476
 477         kmem_cache_free(ceph_inode_cachep, ci);
 478 }
 479
 480 void ceph_destroy_inode(struct inode *inode)
 481 {
 482         struct ceph_inode_info *ci = ceph_inode(inode);
 483         struct ceph_inode_frag *frag;
 484         struct rb_node *n;
 485
 486         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 487
 488         ceph_fscache_unregister_inode_cookie(ci);
 489
 490         ceph_queue_caps_release(inode);
 491
 492         /*
 493          * we may still have a snap_realm reference if there are stray
 494          * caps in i_snap_caps.
 495          */
 496         if (ci->i_snap_realm) {
 497                 struct ceph_mds_client *mdsc =
 498                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 499                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 500
 501                 dout(" dropping residual ref to snap realm %p\n", realm);
 502                 spin_lock(&realm->inodes_with_caps_lock);
 503                 list_del_init(&ci->i_snap_realm_item);
 504                 spin_unlock(&realm->inodes_with_caps_lock);
 505                 ceph_put_snap_realm(mdsc, realm);
 506         }
 507
 508         kfree(ci->i_symlink);
 509         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 510                 frag = rb_entry(n, struct ceph_inode_frag, node);
 511                 rb_erase(n, &ci->i_fragtree);
 512                 kfree(frag);
 513         }
 514
 515         __ceph_destroy_xattrs(ci);
 516         if (ci->i_xattrs.blob)
 517                 ceph_buffer_put(ci->i_xattrs.blob);
 518         if (ci->i_xattrs.prealloc_blob)
 519                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 520
 521         call_rcu(&inode->i_rcu, ceph_i_callback);
 522 }
 523
 524 int ceph_drop_inode(struct inode *inode)
 525 {
 526         /*
 527          * Positve dentry and corresponding inode are always accompanied
 528          * in MDS reply. So no need to keep inode in the cache after
 529          * dropping all its aliases.
 530          */
 531         return 1;
 532 }
 533
 534 /*
 535  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 536  * careful because either the client or MDS may have more up to date
 537  * info, depending on which capabilities are held, and whether
 538  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 539  * and size are monotonically increasing, except when utimes() or
 540  * truncate() increments the corresponding _seq values.)
 541  */
 542 int ceph_fill_file_size(struct inode *inode, int issued,
 543                         u32 truncate_seq, u64 truncate_size, u64 size)
 544 {
 545         struct ceph_inode_info *ci = ceph_inode(inode);
 546         int queue_trunc = 0;
 547
 548         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 549             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 550                 dout("size %lld -> %llu\n", inode->i_size, size);
 551                 inode->i_size = size;
 552                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 553                 ci->i_reported_size = size;
 554                 if (truncate_seq != ci->i_truncate_seq) {
 555                         dout("truncate_seq %u -> %u\n",
 556                              ci->i_truncate_seq, truncate_seq);
 557                         ci->i_truncate_seq = truncate_seq;
 558
 559                         /* the MDS should have revoked these caps */
 560                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 561                                                CEPH_CAP_FILE_RD |
 562                                                CEPH_CAP_FILE_WR |
 563                                                CEPH_CAP_FILE_LAZYIO));
 564                         /*
 565                          * If we hold relevant caps, or in the case where we're
 566                          * not the only client referencing this file and we
 567                          * don't hold those caps, then we need to check whether
 568                          * the file is either opened or mmaped
 569                          */
 570                         if ((issued & (CEPH_CAP_FILE_CACHE|
 571                                        CEPH_CAP_FILE_BUFFER)) ||
 572                             mapping_mapped(inode->i_mapping) ||
 573                             __ceph_caps_file_wanted(ci)) {
 574                                 ci->i_truncate_pending++;
 575                                 queue_trunc = 1;
 576                         }
 577                 }
 578         }
 579         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 580             ci->i_truncate_size != truncate_size) {
 581                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 582                      truncate_size);
 583                 ci->i_truncate_size = truncate_size;
 584         }
 585
 586         if (queue_trunc)
 587                 ceph_fscache_invalidate(inode);
 588
 589         return queue_trunc;
 590 }
 591
 592 void ceph_fill_file_time(struct inode *inode, int issued,
 593                          u64 time_warp_seq, struct timespec *ctime,
 594                          struct timespec *mtime, struct timespec *atime)
 595 {
 596         struct ceph_inode_info *ci = ceph_inode(inode);
 597         int warn = 0;
 598
 599         if (issued & (CEPH_CAP_FILE_EXCL|
 600                       CEPH_CAP_FILE_WR|
 601                       CEPH_CAP_FILE_BUFFER|
 602                       CEPH_CAP_AUTH_EXCL|
 603                       CEPH_CAP_XATTR_EXCL)) {
 604                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 605                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 606                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 607                              ctime->tv_sec, ctime->tv_nsec);
 608                         inode->i_ctime = *ctime;
 609                 }
 610                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 611                         /* the MDS did a utimes() */
 612                         dout("mtime %ld.%09ld -> %ld.%09ld "
 613                              "tw %d -> %d\n",
 614                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 615                              mtime->tv_sec, mtime->tv_nsec,
 616                              ci->i_time_warp_seq, (int)time_warp_seq);
 617
 618                         inode->i_mtime = *mtime;
 619                         inode->i_atime = *atime;
 620                         ci->i_time_warp_seq = time_warp_seq;
 621                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 622                         /* nobody did utimes(); take the max */
 623                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 624                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 625                                      inode->i_mtime.tv_sec,
 626                                      inode->i_mtime.tv_nsec,
 627                                      mtime->tv_sec, mtime->tv_nsec);
 628                                 inode->i_mtime = *mtime;
 629                         }
 630                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 631                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 632                                      inode->i_atime.tv_sec,
 633                                      inode->i_atime.tv_nsec,
 634                                      atime->tv_sec, atime->tv_nsec);
 635                                 inode->i_atime = *atime;
 636                         }
 637                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 638                         /* we did a utimes(); ignore mds values */
 639                 } else {
 640                         warn = 1;
 641                 }
 642         } else {
 643                 /* we have no write|excl caps; whatever the MDS says is true */
 644                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 645                         inode->i_ctime = *ctime;
 646                         inode->i_mtime = *mtime;
 647                         inode->i_atime = *atime;
 648                         ci->i_time_warp_seq = time_warp_seq;
 649                 } else {
 650                         warn = 1;
 651                 }
 652         }
 653         if (warn) /* time_warp_seq shouldn't go backwards */
 654                 dout("%p mds time_warp_seq %llu < %u\n",
 655                      inode, time_warp_seq, ci->i_time_warp_seq);
 656 }
 657
 658 /*
 659  * Populate an inode based on info from mds.  May be called on new or
 660  * existing inodes.
 661  */
 662 static int fill_inode(struct inode *inode, struct page *locked_page,
 663                       struct ceph_mds_reply_info_in *iinfo,
 664                       struct ceph_mds_reply_dirfrag *dirinfo,
 665                       struct ceph_mds_session *session,
 666                       unsigned long ttl_from, int cap_fmode,
 667                       struct ceph_cap_reservation *caps_reservation)
 668 {
 669         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 670         struct ceph_mds_reply_inode *info = iinfo->in;
 671         struct ceph_inode_info *ci = ceph_inode(inode);
 672         int issued = 0, implemented, new_issued;
 673         struct timespec mtime, atime, ctime;
 674         struct ceph_buffer *xattr_blob = NULL;
 675         struct ceph_cap *new_cap = NULL;
 676         int err = 0;
 677         bool wake = false;
 678         bool queue_trunc = false;
 679         bool new_version = false;
 680         bool fill_inline = false;
 681
 682         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 683              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 684              ci->i_version);
 685
 686         /* prealloc new cap struct */
 687         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 688                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 689
 690         /*
 691          * prealloc xattr data, if it looks like we'll need it.  only
 692          * if len > 4 (meaning there are actually xattrs; the first 4
 693          * bytes are the xattr count).
 694          */
 695         if (iinfo->xattr_len > 4) {
 696                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 697                 if (!xattr_blob)
 698                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 699                                iinfo->xattr_len);
 700         }
 701
 702         spin_lock(&ci->i_ceph_lock);
 703
 704         /*
 705          * provided version will be odd if inode value is projected,
 706          * even if stable.  skip the update if we have newer stable
 707          * info (ours>=theirs, e.g. due to racing mds replies), unless
 708          * we are getting projected (unstable) info (in which case the
 709          * version is odd, and we want ours>theirs).
 710          *   us   them
 711          *   2    2     skip
 712          *   3    2     skip
 713          *   3    3     update
 714          */
 715         if (ci->i_version == 0 ||
 716             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 717              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 718                 new_version = true;
 719
 720         issued = __ceph_caps_issued(ci, &implemented);
 721         issued |= implemented | __ceph_caps_dirty(ci);
 722         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 723
 724         /* update inode */
 725         ci->i_version = le64_to_cpu(info->version);
 726         inode->i_version++;
 727         inode->i_rdev = le32_to_cpu(info->rdev);
 728         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 729
 730         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 731             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 732                 inode->i_mode = le32_to_cpu(info->mode);
 733                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 734                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 735                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 736                      from_kuid(&init_user_ns, inode->i_uid),
 737                      from_kgid(&init_user_ns, inode->i_gid));
 738         }
 739
 740         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 741             (issued & CEPH_CAP_LINK_EXCL) == 0)
 742                 set_nlink(inode, le32_to_cpu(info->nlink));
 743
 744         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 745                 /* be careful with mtime, atime, size */
 746                 ceph_decode_timespec(&atime, &info->atime);
 747                 ceph_decode_timespec(&mtime, &info->mtime);
 748                 ceph_decode_timespec(&ctime, &info->ctime);
 749                 ceph_fill_file_time(inode, issued,
 750                                 le32_to_cpu(info->time_warp_seq),
 751                                 &ctime, &mtime, &atime);
 752         }
 753
 754         if (new_version ||
 755             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 756                 ci->i_layout = info->layout;
 757                 queue_trunc = ceph_fill_file_size(inode, issued,
 758                                         le32_to_cpu(info->truncate_seq),
 759                                         le64_to_cpu(info->truncate_size),
 760                                         le64_to_cpu(info->size));
 761                 /* only update max_size on auth cap */
 762                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 763                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 764                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 765                                         le64_to_cpu(info->max_size));
 766                         ci->i_max_size = le64_to_cpu(info->max_size);
 767                 }
 768         }
 769
 770         /* xattrs */
 771         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 772         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 773             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 774                 if (ci->i_xattrs.blob)
 775                         ceph_buffer_put(ci->i_xattrs.blob);
 776                 ci->i_xattrs.blob = xattr_blob;
 777                 if (xattr_blob)
 778                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 779                                iinfo->xattr_data, iinfo->xattr_len);
 780                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 781                 ceph_forget_all_cached_acls(inode);
 782                 xattr_blob = NULL;
 783         }
 784
 785         inode->i_mapping->a_ops = &ceph_aops;
 786
 787         switch (inode->i_mode & S_IFMT) {
 788         case S_IFIFO:
 789         case S_IFBLK:
 790         case S_IFCHR:
 791         case S_IFSOCK:
 792                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 793                 inode->i_op = &ceph_file_iops;
 794                 break;
 795         case S_IFREG:
 796                 inode->i_op = &ceph_file_iops;
 797                 inode->i_fop = &ceph_file_fops;
 798                 break;
 799         case S_IFLNK:
 800                 inode->i_op = &ceph_symlink_iops;
 801                 if (!ci->i_symlink) {
 802                         u32 symlen = iinfo->symlink_len;
 803                         char *sym;
 804
 805                         spin_unlock(&ci->i_ceph_lock);
 806
 807                         err = -EINVAL;
 808                         if (WARN_ON(symlen != inode->i_size))
 809                                 goto out;
 810
 811                         err = -ENOMEM;
 812                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 813                         if (!sym)
 814                                 goto out;
 815
 816                         spin_lock(&ci->i_ceph_lock);
 817                         if (!ci->i_symlink)
 818                                 ci->i_symlink = sym;
 819                         else
 820                                 kfree(sym); /* lost a race */
 821                 }
 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         /* set dir completion flag? */
 842         if (S_ISDIR(inode->i_mode) &&
 843             ci->i_files == 0 && ci->i_subdirs == 0 &&
 844             ceph_snap(inode) == CEPH_NOSNAP &&
 845             (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
 846             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 847             !__ceph_dir_is_complete(ci)) {
 848                 dout(" marking %p complete (empty)\n", inode);
 849                 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count),
 850                                         ci->i_ordered_count);
 851         }
 852
 853         /* were we issued a capability? */
 854         if (info->cap.caps) {
 855                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 856                         ceph_add_cap(inode, session,
 857                                      le64_to_cpu(info->cap.cap_id),
 858                                      cap_fmode,
 859                                      le32_to_cpu(info->cap.caps),
 860                                      le32_to_cpu(info->cap.wanted),
 861                                      le32_to_cpu(info->cap.seq),
 862                                      le32_to_cpu(info->cap.mseq),
 863                                      le64_to_cpu(info->cap.realm),
 864                                      info->cap.flags, &new_cap);
 865                         wake = true;
 866                 } else {
 867                         dout(" %p got snap_caps %s\n", inode,
 868                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 869                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 870                         if (cap_fmode >= 0)
 871                                 __ceph_get_fmode(ci, cap_fmode);
 872                 }
 873         } else if (cap_fmode >= 0) {
 874                 pr_warn("mds issued no caps on %llx.%llx\n",
 875                            ceph_vinop(inode));
 876                 __ceph_get_fmode(ci, cap_fmode);
 877         }
 878
 879         if (iinfo->inline_version > 0 &&
 880             iinfo->inline_version >= ci->i_inline_version) {
 881                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 882                 ci->i_inline_version = iinfo->inline_version;
 883                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 884                     (locked_page ||
 885                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 886                         fill_inline = true;
 887         }
 888
 889         spin_unlock(&ci->i_ceph_lock);
 890
 891         if (fill_inline)
 892                 ceph_fill_inline_data(inode, locked_page,
 893                                       iinfo->inline_data, iinfo->inline_len);
 894
 895         if (wake)
 896                 wake_up_all(&ci->i_cap_wq);
 897
 898         /* queue truncate if we saw i_size decrease */
 899         if (queue_trunc)
 900                 ceph_queue_vmtruncate(inode);
 901
 902         /* populate frag tree */
 903         if (S_ISDIR(inode->i_mode))
 904                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 905
 906         /* update delegation info? */
 907         if (dirinfo)
 908                 ceph_fill_dirfrag(inode, dirinfo);
 909
 910         err = 0;
 911 out:
 912         if (new_cap)
 913                 ceph_put_cap(mdsc, new_cap);
 914         if (xattr_blob)
 915                 ceph_buffer_put(xattr_blob);
 916         return err;
 917 }
 918
 919 /*
 920  * caller should hold session s_mutex.
 921  */
 922 static void update_dentry_lease(struct dentry *dentry,
 923                                 struct ceph_mds_reply_lease *lease,
 924                                 struct ceph_mds_session *session,
 925                                 unsigned long from_time)
 926 {
 927         struct ceph_dentry_info *di = ceph_dentry(dentry);
 928         long unsigned duration = le32_to_cpu(lease->duration_ms);
 929         long unsigned ttl = from_time + (duration * HZ) / 1000;
 930         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 931         struct inode *dir;
 932
 933         /* only track leases on regular dentries */
 934         if (dentry->d_op != &ceph_dentry_ops)
 935                 return;
 936
 937         spin_lock(&dentry->d_lock);
 938         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 939              dentry, duration, ttl);
 940
 941         /* make lease_rdcache_gen match directory */
 942         dir = dentry->d_parent->d_inode;
 943         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 944
 945         if (duration == 0)
 946                 goto out_unlock;
 947
 948         if (di->lease_gen == session->s_cap_gen &&
 949             time_before(ttl, dentry->d_time))
 950                 goto out_unlock;  /* we already have a newer lease. */
 951
 952         if (di->lease_session && di->lease_session != session)
 953                 goto out_unlock;
 954
 955         ceph_dentry_lru_touch(dentry);
 956
 957         if (!di->lease_session)
 958                 di->lease_session = ceph_get_mds_session(session);
 959         di->lease_gen = session->s_cap_gen;
 960         di->lease_seq = le32_to_cpu(lease->seq);
 961         di->lease_renew_after = half_ttl;
 962         di->lease_renew_from = 0;
 963         dentry->d_time = ttl;
 964 out_unlock:
 965         spin_unlock(&dentry->d_lock);
 966         return;
 967 }
 968
 969 /*
 970  * splice a dentry to an inode.
 971  * caller must hold directory i_mutex for this to be safe.
 972  *
 973  * we will only rehash the resulting dentry if @prehash is
 974  * true; @prehash will be set to false (for the benefit of
 975  * the caller) if we fail.
 976  */
 977 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 978                                     bool *prehash)
 979 {
 980         struct dentry *realdn;
 981
 982         BUG_ON(dn->d_inode);
 983
 984         /* dn must be unhashed */
 985         if (!d_unhashed(dn))
 986                 d_drop(dn);
 987         realdn = d_splice_alias(in, dn);
 988         if (IS_ERR(realdn)) {
 989                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 990                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 991                 if (prehash)
 992                         *prehash = false; /* don't rehash on error */
 993                 dn = realdn; /* note realdn contains the error */
 994                 goto out;
 995         } else if (realdn) {
 996                 dout("dn %p (%d) spliced with %p (%d) "
 997                      "inode %p ino %llx.%llx\n",
 998                      dn, d_count(dn),
 999                      realdn, d_count(realdn),
1000                      realdn->d_inode, ceph_vinop(realdn->d_inode));
1001                 dput(dn);
1002                 dn = realdn;
1003         } else {
1004                 BUG_ON(!ceph_dentry(dn));
1005                 dout("dn %p attached to %p ino %llx.%llx\n",
1006                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
1007         }
1008         if ((!prehash || *prehash) && d_unhashed(dn))
1009                 d_rehash(dn);
1010 out:
1011         return dn;
1012 }
1013
1014 /*
1015  * Incorporate results into the local cache.  This is either just
1016  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1017  * after a lookup).
1018  *
1019  * A reply may contain
1020  *         a directory inode along with a dentry.
1021  *  and/or a target inode
1022  *
1023  * Called with snap_rwsem (read).
1024  */
1025 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1026                     struct ceph_mds_session *session)
1027 {
1028         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1029         struct inode *in = NULL;
1030         struct ceph_vino vino;
1031         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1032         int err = 0;
1033
1034         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1035              rinfo->head->is_dentry, rinfo->head->is_target);
1036
1037 #if 0
1038         /*
1039          * Debugging hook:
1040          *
1041          * If we resend completed ops to a recovering mds, we get no
1042          * trace.  Since that is very rare, pretend this is the case
1043          * to ensure the 'no trace' handlers in the callers behave.
1044          *
1045          * Fill in inodes unconditionally to avoid breaking cap
1046          * invariants.
1047          */
1048         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1049                 pr_info("fill_trace faking empty trace on %lld %s\n",
1050                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1051                 if (rinfo->head->is_dentry) {
1052                         rinfo->head->is_dentry = 0;
1053                         err = fill_inode(req->r_locked_dir,
1054                                          &rinfo->diri, rinfo->dirfrag,
1055                                          session, req->r_request_started, -1);
1056                 }
1057                 if (rinfo->head->is_target) {
1058                         rinfo->head->is_target = 0;
1059                         ininfo = rinfo->targeti.in;
1060                         vino.ino = le64_to_cpu(ininfo->ino);
1061                         vino.snap = le64_to_cpu(ininfo->snapid);
1062                         in = ceph_get_inode(sb, vino);
1063                         err = fill_inode(in, &rinfo->targeti, NULL,
1064                                          session, req->r_request_started,
1065                                          req->r_fmode);
1066                         iput(in);
1067                 }
1068         }
1069 #endif
1070
1071         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1072                 dout("fill_trace reply is empty!\n");
1073                 if (rinfo->head->result == 0 && req->r_locked_dir)
1074                         ceph_invalidate_dir_request(req);
1075                 return 0;
1076         }
1077
1078         if (rinfo->head->is_dentry) {
1079                 struct inode *dir = req->r_locked_dir;
1080
1081                 if (dir) {
1082                         err = fill_inode(dir, NULL,
1083                                          &rinfo->diri, rinfo->dirfrag,
1084                                          session, req->r_request_started, -1,
1085                                          &req->r_caps_reservation);
1086                         if (err < 0)
1087                                 goto done;
1088                 } else {
1089                         WARN_ON_ONCE(1);
1090                 }
1091
1092                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1093                         struct qstr dname;
1094                         struct dentry *dn, *parent;
1095
1096                         BUG_ON(!rinfo->head->is_target);
1097                         BUG_ON(req->r_dentry);
1098
1099                         parent = d_find_any_alias(dir);
1100                         BUG_ON(!parent);
1101
1102                         dname.name = rinfo->dname;
1103                         dname.len = rinfo->dname_len;
1104                         dname.hash = full_name_hash(dname.name, dname.len);
1105                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1106                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1107 retry_lookup:
1108                         dn = d_lookup(parent, &dname);
1109                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1110                              parent, dname.len, dname.name, dn);
1111
1112                         if (!dn) {
1113                                 dn = d_alloc(parent, &dname);
1114                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1115                                      dname.len, dname.name, dn);
1116                                 if (dn == NULL) {
1117                                         dput(parent);
1118                                         err = -ENOMEM;
1119                                         goto done;
1120                                 }
1121                                 err = ceph_init_dentry(dn);
1122                                 if (err < 0) {
1123                                         dput(dn);
1124                                         dput(parent);
1125                                         goto done;
1126                                 }
1127                         } else if (dn->d_inode &&
1128                                    (ceph_ino(dn->d_inode) != vino.ino ||
1129                                     ceph_snap(dn->d_inode) != vino.snap)) {
1130                                 dout(" dn %p points to wrong inode %p\n",
1131                                      dn, dn->d_inode);
1132                                 d_delete(dn);
1133                                 dput(dn);
1134                                 goto retry_lookup;
1135                         }
1136
1137                         req->r_dentry = dn;
1138                         dput(parent);
1139                 }
1140         }
1141
1142         if (rinfo->head->is_target) {
1143                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1144                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1145
1146                 in = ceph_get_inode(sb, vino);
1147                 if (IS_ERR(in)) {
1148                         err = PTR_ERR(in);
1149                         goto done;
1150                 }
1151                 req->r_target_inode = in;
1152
1153                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1154                                 session, req->r_request_started,
1155                                 (!req->r_aborted && rinfo->head->result == 0) ?
1156                                 req->r_fmode : -1,
1157                                 &req->r_caps_reservation);
1158                 if (err < 0) {
1159                         pr_err("fill_inode badness %p %llx.%llx\n",
1160                                 in, ceph_vinop(in));
1161                         goto done;
1162                 }
1163         }
1164
1165         /*
1166          * ignore null lease/binding on snapdir ENOENT, or else we
1167          * will have trouble splicing in the virtual snapdir later
1168          */
1169         if (rinfo->head->is_dentry && !req->r_aborted &&
1170             req->r_locked_dir &&
1171             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1172                                                fsc->mount_options->snapdir_name,
1173                                                req->r_dentry->d_name.len))) {
1174                 /*
1175                  * lookup link rename   : null -> possibly existing inode
1176                  * mknod symlink mkdir  : null -> new inode
1177                  * unlink               : linked -> null
1178                  */
1179                 struct inode *dir = req->r_locked_dir;
1180                 struct dentry *dn = req->r_dentry;
1181                 bool have_dir_cap, have_lease;
1182
1183                 BUG_ON(!dn);
1184                 BUG_ON(!dir);
1185                 BUG_ON(dn->d_parent->d_inode != dir);
1186                 BUG_ON(ceph_ino(dir) !=
1187                        le64_to_cpu(rinfo->diri.in->ino));
1188                 BUG_ON(ceph_snap(dir) !=
1189                        le64_to_cpu(rinfo->diri.in->snapid));
1190
1191                 /* do we have a lease on the whole dir? */
1192                 have_dir_cap =
1193                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1194                          CEPH_CAP_FILE_SHARED);
1195
1196                 /* do we have a dn lease? */
1197                 have_lease = have_dir_cap ||
1198                         le32_to_cpu(rinfo->dlease->duration_ms);
1199                 if (!have_lease)
1200                         dout("fill_trace  no dentry lease or dir cap\n");
1201
1202                 /* rename? */
1203                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1204                         struct inode *olddir = req->r_old_dentry_dir;
1205                         BUG_ON(!olddir);
1206
1207                         dout(" src %p '%pd' dst %p '%pd'\n",
1208                              req->r_old_dentry,
1209                              req->r_old_dentry,
1210                              dn, dn);
1211                         dout("fill_trace doing d_move %p -> %p\n",
1212                              req->r_old_dentry, dn);
1213
1214                         d_move(req->r_old_dentry, dn);
1215                         dout(" src %p '%pd' dst %p '%pd'\n",
1216                              req->r_old_dentry,
1217                              req->r_old_dentry,
1218                              dn, dn);
1219
1220                         /* ensure target dentry is invalidated, despite
1221                            rehashing bug in vfs_rename_dir */
1222                         ceph_invalidate_dentry_lease(dn);
1223
1224                         /* d_move screws up sibling dentries' offsets */
1225                         ceph_dir_clear_ordered(dir);
1226                         ceph_dir_clear_ordered(olddir);
1227
1228                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1229                              ceph_dentry(req->r_old_dentry)->offset);
1230
1231                         dn = req->r_old_dentry;  /* use old_dentry */
1232                 }
1233
1234                 /* null dentry? */
1235                 if (!rinfo->head->is_target) {
1236                         dout("fill_trace null dentry\n");
1237                         if (dn->d_inode) {
1238                                 ceph_dir_clear_ordered(dir);
1239                                 dout("d_delete %p\n", dn);
1240                                 d_delete(dn);
1241                         } else {
1242                                 dout("d_instantiate %p NULL\n", dn);
1243                                 d_instantiate(dn, NULL);
1244                                 if (have_lease && d_unhashed(dn))
1245                                         d_rehash(dn);
1246                                 update_dentry_lease(dn, rinfo->dlease,
1247                                                     session,
1248                                                     req->r_request_started);
1249                         }
1250                         goto done;
1251                 }
1252
1253                 /* attach proper inode */
1254                 if (!dn->d_inode) {
1255                         ceph_dir_clear_ordered(dir);
1256                         ihold(in);
1257                         dn = splice_dentry(dn, in, &have_lease);
1258                         if (IS_ERR(dn)) {
1259                                 err = PTR_ERR(dn);
1260                                 goto done;
1261                         }
1262                         req->r_dentry = dn;  /* may have spliced */
1263                 } else if (dn->d_inode && dn->d_inode != in) {
1264                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1265                              dn, dn->d_inode, ceph_vinop(dn->d_inode),
1266                              ceph_vinop(in));
1267                         have_lease = false;
1268                 }
1269
1270                 if (have_lease)
1271                         update_dentry_lease(dn, rinfo->dlease, session,
1272                                             req->r_request_started);
1273                 dout(" final dn %p\n", dn);
1274         } else if (!req->r_aborted &&
1275                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1276                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1277                 struct dentry *dn = req->r_dentry;
1278                 struct inode *dir = req->r_locked_dir;
1279
1280                 /* fill out a snapdir LOOKUPSNAP dentry */
1281                 BUG_ON(!dn);
1282                 BUG_ON(!dir);
1283                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1284                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1285                 ceph_dir_clear_ordered(dir);
1286                 ihold(in);
1287                 dn = splice_dentry(dn, in, NULL);
1288                 if (IS_ERR(dn)) {
1289                         err = PTR_ERR(dn);
1290                         goto done;
1291                 }
1292                 req->r_dentry = dn;  /* may have spliced */
1293         }
1294 done:
1295         dout("fill_trace done err=%d\n", err);
1296         return err;
1297 }
1298
1299 /*
1300  * Prepopulate our cache with readdir results, leases, etc.
1301  */
1302 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1303                                            struct ceph_mds_session *session)
1304 {
1305         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1306         int i, err = 0;
1307
1308         for (i = 0; i < rinfo->dir_nr; i++) {
1309                 struct ceph_vino vino;
1310                 struct inode *in;
1311                 int rc;
1312
1313                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1314                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1315
1316                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1317                 if (IS_ERR(in)) {
1318                         err = PTR_ERR(in);
1319                         dout("new_inode badness got %d\n", err);
1320                         continue;
1321                 }
1322                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1323                                 req->r_request_started, -1,
1324                                 &req->r_caps_reservation);
1325                 if (rc < 0) {
1326                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1327                         err = rc;
1328                         continue;
1329                 }
1330         }
1331
1332         return err;
1333 }
1334
1335 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1336                              struct ceph_mds_session *session)
1337 {
1338         struct dentry *parent = req->r_dentry;
1339         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1340         struct qstr dname;
1341         struct dentry *dn;
1342         struct inode *in;
1343         int err = 0, ret, i;
1344         struct inode *snapdir = NULL;
1345         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1346         struct ceph_dentry_info *di;
1347         u64 r_readdir_offset = req->r_readdir_offset;
1348         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1349
1350         if (rinfo->dir_dir &&
1351             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1352                 dout("readdir_prepopulate got new frag %x -> %x\n",
1353                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1354                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1355                 if (ceph_frag_is_leftmost(frag))
1356                         r_readdir_offset = 2;
1357                 else
1358                         r_readdir_offset = 0;
1359         }
1360
1361         if (req->r_aborted)
1362                 return readdir_prepopulate_inodes_only(req, session);
1363
1364         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1365                 snapdir = ceph_get_snapdir(parent->d_inode);
1366                 parent = d_find_alias(snapdir);
1367                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1368                      rinfo->dir_nr, parent);
1369         } else {
1370                 dout("readdir_prepopulate %d items under dn %p\n",
1371                      rinfo->dir_nr, parent);
1372                 if (rinfo->dir_dir)
1373                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1374         }
1375
1376         /* FIXME: release caps/leases if error occurs */
1377         for (i = 0; i < rinfo->dir_nr; i++) {
1378                 struct ceph_vino vino;
1379
1380                 dname.name = rinfo->dir_dname[i];
1381                 dname.len = rinfo->dir_dname_len[i];
1382                 dname.hash = full_name_hash(dname.name, dname.len);
1383
1384                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1385                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1386
1387 retry_lookup:
1388                 dn = d_lookup(parent, &dname);
1389                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1390                      parent, dname.len, dname.name, dn);
1391
1392                 if (!dn) {
1393                         dn = d_alloc(parent, &dname);
1394                         dout("d_alloc %p '%.*s' = %p\n", parent,
1395                              dname.len, dname.name, dn);
1396                         if (dn == NULL) {
1397                                 dout("d_alloc badness\n");
1398                                 err = -ENOMEM;
1399                                 goto out;
1400                         }
1401                         ret = ceph_init_dentry(dn);
1402                         if (ret < 0) {
1403                                 dput(dn);
1404                                 err = ret;
1405                                 goto out;
1406                         }
1407                 } else if (dn->d_inode &&
1408                            (ceph_ino(dn->d_inode) != vino.ino ||
1409                             ceph_snap(dn->d_inode) != vino.snap)) {
1410                         dout(" dn %p points to wrong inode %p\n",
1411                              dn, dn->d_inode);
1412                         d_delete(dn);
1413                         dput(dn);
1414                         goto retry_lookup;
1415                 } else {
1416                         /* reorder parent's d_subdirs */
1417                         spin_lock(&parent->d_lock);
1418                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1419                         list_move(&dn->d_child, &parent->d_subdirs);
1420                         spin_unlock(&dn->d_lock);
1421                         spin_unlock(&parent->d_lock);
1422                 }
1423
1424                 /* inode */
1425                 if (dn->d_inode) {
1426                         in = dn->d_inode;
1427                 } else {
1428                         in = ceph_get_inode(parent->d_sb, vino);
1429                         if (IS_ERR(in)) {
1430                                 dout("new_inode badness\n");
1431                                 d_drop(dn);
1432                                 dput(dn);
1433                                 err = PTR_ERR(in);
1434                                 goto out;
1435                         }
1436                 }
1437
1438                 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1439                                req->r_request_started, -1,
1440                                &req->r_caps_reservation) < 0) {
1441                         pr_err("fill_inode badness on %p\n", in);
1442                         if (!dn->d_inode)
1443                                 iput(in);
1444                         d_drop(dn);
1445                         goto next_item;
1446                 }
1447
1448                 if (!dn->d_inode) {
1449                         dn = splice_dentry(dn, in, NULL);
1450                         if (IS_ERR(dn)) {
1451                                 err = PTR_ERR(dn);
1452                                 dn = NULL;
1453                                 goto next_item;
1454                         }
1455                 }
1456
1457                 di = dn->d_fsdata;
1458                 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1459
1460                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1461                                     req->r_session,
1462                                     req->r_request_started);
1463 next_item:
1464                 if (dn)
1465                         dput(dn);
1466         }
1467         if (err == 0)
1468                 req->r_did_prepopulate = true;
1469
1470 out:
1471         if (snapdir) {
1472                 iput(snapdir);
1473                 dput(parent);
1474         }
1475         dout("readdir_prepopulate done\n");
1476         return err;
1477 }
1478
1479 int ceph_inode_set_size(struct inode *inode, loff_t size)
1480 {
1481         struct ceph_inode_info *ci = ceph_inode(inode);
1482         int ret = 0;
1483
1484         spin_lock(&ci->i_ceph_lock);
1485         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1486         inode->i_size = size;
1487         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1488
1489         /* tell the MDS if we are approaching max_size */
1490         if ((size << 1) >= ci->i_max_size &&
1491             (ci->i_reported_size << 1) < ci->i_max_size)
1492                 ret = 1;
1493
1494         spin_unlock(&ci->i_ceph_lock);
1495         return ret;
1496 }
1497
1498 /*
1499  * Write back inode data in a worker thread.  (This can't be done
1500  * in the message handler context.)
1501  */
1502 void ceph_queue_writeback(struct inode *inode)
1503 {
1504         ihold(inode);
1505         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1506                        &ceph_inode(inode)->i_wb_work)) {
1507                 dout("ceph_queue_writeback %p\n", inode);
1508         } else {
1509                 dout("ceph_queue_writeback %p failed\n", inode);
1510                 iput(inode);
1511         }
1512 }
1513
1514 static void ceph_writeback_work(struct work_struct *work)
1515 {
1516         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1517                                                   i_wb_work);
1518         struct inode *inode = &ci->vfs_inode;
1519
1520         dout("writeback %p\n", inode);
1521         filemap_fdatawrite(&inode->i_data);
1522         iput(inode);
1523 }
1524
1525 /*
1526  * queue an async invalidation
1527  */
1528 void ceph_queue_invalidate(struct inode *inode)
1529 {
1530         ihold(inode);
1531         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1532                        &ceph_inode(inode)->i_pg_inv_work)) {
1533                 dout("ceph_queue_invalidate %p\n", inode);
1534         } else {
1535                 dout("ceph_queue_invalidate %p failed\n", inode);
1536                 iput(inode);
1537         }
1538 }
1539
1540 /*
1541  * Invalidate inode pages in a worker thread.  (This can't be done
1542  * in the message handler context.)
1543  */
1544 static void ceph_invalidate_work(struct work_struct *work)
1545 {
1546         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1547                                                   i_pg_inv_work);
1548         struct inode *inode = &ci->vfs_inode;
1549         u32 orig_gen;
1550         int check = 0;
1551
1552         mutex_lock(&ci->i_truncate_mutex);
1553         spin_lock(&ci->i_ceph_lock);
1554         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1555              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1556         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1557                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1558                         check = 1;
1559                 spin_unlock(&ci->i_ceph_lock);
1560                 mutex_unlock(&ci->i_truncate_mutex);
1561                 goto out;
1562         }
1563         orig_gen = ci->i_rdcache_gen;
1564         spin_unlock(&ci->i_ceph_lock);
1565
1566         truncate_pagecache(inode, 0);
1567
1568         spin_lock(&ci->i_ceph_lock);
1569         if (orig_gen == ci->i_rdcache_gen &&
1570             orig_gen == ci->i_rdcache_revoking) {
1571                 dout("invalidate_pages %p gen %d successful\n", inode,
1572                      ci->i_rdcache_gen);
1573                 ci->i_rdcache_revoking--;
1574                 check = 1;
1575         } else {
1576                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1577                      inode, orig_gen, ci->i_rdcache_gen,
1578                      ci->i_rdcache_revoking);
1579                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1580                         check = 1;
1581         }
1582         spin_unlock(&ci->i_ceph_lock);
1583         mutex_unlock(&ci->i_truncate_mutex);
1584 out:
1585         if (check)
1586                 ceph_check_caps(ci, 0, NULL);
1587         iput(inode);
1588 }
1589
1590
1591 /*
1592  * called by trunc_wq;
1593  *
1594  * We also truncate in a separate thread as well.
1595  */
1596 static void ceph_vmtruncate_work(struct work_struct *work)
1597 {
1598         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1599                                                   i_vmtruncate_work);
1600         struct inode *inode = &ci->vfs_inode;
1601
1602         dout("vmtruncate_work %p\n", inode);
1603         __ceph_do_pending_vmtruncate(inode);
1604         iput(inode);
1605 }
1606
1607 /*
1608  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1609  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1610  */
1611 void ceph_queue_vmtruncate(struct inode *inode)
1612 {
1613         struct ceph_inode_info *ci = ceph_inode(inode);
1614
1615         ihold(inode);
1616
1617         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1618                        &ci->i_vmtruncate_work)) {
1619                 dout("ceph_queue_vmtruncate %p\n", inode);
1620         } else {
1621                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1622                      inode, ci->i_truncate_pending);
1623                 iput(inode);
1624         }
1625 }
1626
1627 /*
1628  * Make sure any pending truncation is applied before doing anything
1629  * that may depend on it.
1630  */
1631 void __ceph_do_pending_vmtruncate(struct inode *inode)
1632 {
1633         struct ceph_inode_info *ci = ceph_inode(inode);
1634         u64 to;
1635         int wrbuffer_refs, finish = 0;
1636
1637         mutex_lock(&ci->i_truncate_mutex);
1638 retry:
1639         spin_lock(&ci->i_ceph_lock);
1640         if (ci->i_truncate_pending == 0) {
1641                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1642                 spin_unlock(&ci->i_ceph_lock);
1643                 mutex_unlock(&ci->i_truncate_mutex);
1644                 return;
1645         }
1646
1647         /*
1648          * make sure any dirty snapped pages are flushed before we
1649          * possibly truncate them.. so write AND block!
1650          */
1651         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1652                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1653                      inode);
1654                 spin_unlock(&ci->i_ceph_lock);
1655                 filemap_write_and_wait_range(&inode->i_data, 0,
1656                                              inode->i_sb->s_maxbytes);
1657                 goto retry;
1658         }
1659
1660         /* there should be no reader or writer */
1661         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1662
1663         to = ci->i_truncate_size;
1664         wrbuffer_refs = ci->i_wrbuffer_ref;
1665         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1666              ci->i_truncate_pending, to);
1667         spin_unlock(&ci->i_ceph_lock);
1668
1669         truncate_pagecache(inode, to);
1670
1671         spin_lock(&ci->i_ceph_lock);
1672         if (to == ci->i_truncate_size) {
1673                 ci->i_truncate_pending = 0;
1674                 finish = 1;
1675         }
1676         spin_unlock(&ci->i_ceph_lock);
1677         if (!finish)
1678                 goto retry;
1679
1680         mutex_unlock(&ci->i_truncate_mutex);
1681
1682         if (wrbuffer_refs == 0)
1683                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1684
1685         wake_up_all(&ci->i_cap_wq);
1686 }
1687
1688 /*
1689  * symlinks
1690  */
1691 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1692 {
1693         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1694         nd_set_link(nd, ci->i_symlink);
1695         return NULL;
1696 }
1697
1698 static const struct inode_operations ceph_symlink_iops = {
1699         .readlink = generic_readlink,
1700         .follow_link = ceph_sym_follow_link,
1701         .setattr = ceph_setattr,
1702         .getattr = ceph_getattr,
1703         .setxattr = ceph_setxattr,
1704         .getxattr = ceph_getxattr,
1705         .listxattr = ceph_listxattr,
1706         .removexattr = ceph_removexattr,
1707 };
1708
1709 /*
1710  * setattr
1711  */
1712 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1713 {
1714         struct inode *inode = dentry->d_inode;
1715         struct ceph_inode_info *ci = ceph_inode(inode);
1716         const unsigned int ia_valid = attr->ia_valid;
1717         struct ceph_mds_request *req;
1718         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1719         int issued;
1720         int release = 0, dirtied = 0;
1721         int mask = 0;
1722         int err = 0;
1723         int inode_dirty_flags = 0;
1724
1725         if (ceph_snap(inode) != CEPH_NOSNAP)
1726                 return -EROFS;
1727
1728         err = inode_change_ok(inode, attr);
1729         if (err != 0)
1730                 return err;
1731
1732         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1733                                        USE_AUTH_MDS);
1734         if (IS_ERR(req))
1735                 return PTR_ERR(req);
1736
1737         spin_lock(&ci->i_ceph_lock);
1738         issued = __ceph_caps_issued(ci, NULL);
1739         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1740
1741         if (ia_valid & ATTR_UID) {
1742                 dout("setattr %p uid %d -> %d\n", inode,
1743                      from_kuid(&init_user_ns, inode->i_uid),
1744                      from_kuid(&init_user_ns, attr->ia_uid));
1745                 if (issued & CEPH_CAP_AUTH_EXCL) {
1746                         inode->i_uid = attr->ia_uid;
1747                         dirtied |= CEPH_CAP_AUTH_EXCL;
1748                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1749                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1750                         req->r_args.setattr.uid = cpu_to_le32(
1751                                 from_kuid(&init_user_ns, attr->ia_uid));
1752                         mask |= CEPH_SETATTR_UID;
1753                         release |= CEPH_CAP_AUTH_SHARED;
1754                 }
1755         }
1756         if (ia_valid & ATTR_GID) {
1757                 dout("setattr %p gid %d -> %d\n", inode,
1758                      from_kgid(&init_user_ns, inode->i_gid),
1759                      from_kgid(&init_user_ns, attr->ia_gid));
1760                 if (issued & CEPH_CAP_AUTH_EXCL) {
1761                         inode->i_gid = attr->ia_gid;
1762                         dirtied |= CEPH_CAP_AUTH_EXCL;
1763                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1764                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1765                         req->r_args.setattr.gid = cpu_to_le32(
1766                                 from_kgid(&init_user_ns, attr->ia_gid));
1767                         mask |= CEPH_SETATTR_GID;
1768                         release |= CEPH_CAP_AUTH_SHARED;
1769                 }
1770         }
1771         if (ia_valid & ATTR_MODE) {
1772                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1773                      attr->ia_mode);
1774                 if (issued & CEPH_CAP_AUTH_EXCL) {
1775                         inode->i_mode = attr->ia_mode;
1776                         dirtied |= CEPH_CAP_AUTH_EXCL;
1777                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1778                            attr->ia_mode != inode->i_mode) {
1779                         inode->i_mode = attr->ia_mode;
1780                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1781                         mask |= CEPH_SETATTR_MODE;
1782                         release |= CEPH_CAP_AUTH_SHARED;
1783                 }
1784         }
1785
1786         if (ia_valid & ATTR_ATIME) {
1787                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1788                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1789                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1790                 if (issued & CEPH_CAP_FILE_EXCL) {
1791                         ci->i_time_warp_seq++;
1792                         inode->i_atime = attr->ia_atime;
1793                         dirtied |= CEPH_CAP_FILE_EXCL;
1794                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1795                            timespec_compare(&inode->i_atime,
1796                                             &attr->ia_atime) < 0) {
1797                         inode->i_atime = attr->ia_atime;
1798                         dirtied |= CEPH_CAP_FILE_WR;
1799                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1800                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1801                         ceph_encode_timespec(&req->r_args.setattr.atime,
1802                                              &attr->ia_atime);
1803                         mask |= CEPH_SETATTR_ATIME;
1804                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1805                                 CEPH_CAP_FILE_WR;
1806                 }
1807         }
1808         if (ia_valid & ATTR_MTIME) {
1809                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1810                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1811                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1812                 if (issued & CEPH_CAP_FILE_EXCL) {
1813                         ci->i_time_warp_seq++;
1814                         inode->i_mtime = attr->ia_mtime;
1815                         dirtied |= CEPH_CAP_FILE_EXCL;
1816                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1817                            timespec_compare(&inode->i_mtime,
1818                                             &attr->ia_mtime) < 0) {
1819                         inode->i_mtime = attr->ia_mtime;
1820                         dirtied |= CEPH_CAP_FILE_WR;
1821                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1822                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1823                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1824                                              &attr->ia_mtime);
1825                         mask |= CEPH_SETATTR_MTIME;
1826                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1827                                 CEPH_CAP_FILE_WR;
1828                 }
1829         }
1830         if (ia_valid & ATTR_SIZE) {
1831                 dout("setattr %p size %lld -> %lld\n", inode,
1832                      inode->i_size, attr->ia_size);
1833                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1834                     attr->ia_size > inode->i_size) {
1835                         inode->i_size = attr->ia_size;
1836                         inode->i_blocks =
1837                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1838                         inode->i_ctime = attr->ia_ctime;
1839                         ci->i_reported_size = attr->ia_size;
1840                         dirtied |= CEPH_CAP_FILE_EXCL;
1841                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1842                            attr->ia_size != inode->i_size) {
1843                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1844                         req->r_args.setattr.old_size =
1845                                 cpu_to_le64(inode->i_size);
1846                         mask |= CEPH_SETATTR_SIZE;
1847                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1848                                 CEPH_CAP_FILE_WR;
1849                 }
1850         }
1851
1852         /* these do nothing */
1853         if (ia_valid & ATTR_CTIME) {
1854                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1855                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1856                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1857                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1858                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1859                      only ? "ctime only" : "ignored");
1860                 inode->i_ctime = attr->ia_ctime;
1861                 if (only) {
1862                         /*
1863                          * if kernel wants to dirty ctime but nothing else,
1864                          * we need to choose a cap to dirty under, or do
1865                          * a almost-no-op setattr
1866                          */
1867                         if (issued & CEPH_CAP_AUTH_EXCL)
1868                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1869                         else if (issued & CEPH_CAP_FILE_EXCL)
1870                                 dirtied |= CEPH_CAP_FILE_EXCL;
1871                         else if (issued & CEPH_CAP_XATTR_EXCL)
1872                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1873                         else
1874                                 mask |= CEPH_SETATTR_CTIME;
1875                 }
1876         }
1877         if (ia_valid & ATTR_FILE)
1878                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1879
1880         if (dirtied) {
1881                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1882                 inode->i_ctime = CURRENT_TIME;
1883         }
1884
1885         release &= issued;
1886         spin_unlock(&ci->i_ceph_lock);
1887
1888         if (inode_dirty_flags)
1889                 __mark_inode_dirty(inode, inode_dirty_flags);
1890
1891         if (ia_valid & ATTR_MODE) {
1892                 err = posix_acl_chmod(inode, attr->ia_mode);
1893                 if (err)
1894                         goto out_put;
1895         }
1896
1897         if (mask) {
1898                 req->r_inode = inode;
1899                 ihold(inode);
1900                 req->r_inode_drop = release;
1901                 req->r_args.setattr.mask = cpu_to_le32(mask);
1902                 req->r_num_caps = 1;
1903                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1904         }
1905         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1906              ceph_cap_string(dirtied), mask);
1907
1908         ceph_mdsc_put_request(req);
1909         if (mask & CEPH_SETATTR_SIZE)
1910                 __ceph_do_pending_vmtruncate(inode);
1911         return err;
1912 out_put:
1913         ceph_mdsc_put_request(req);
1914         return err;
1915 }
1916
1917 /*
1918  * Verify that we have a lease on the given mask.  If not,
1919  * do a getattr against an mds.
1920  */
1921 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1922                       int mask, bool force)
1923 {
1924         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1925         struct ceph_mds_client *mdsc = fsc->mdsc;
1926         struct ceph_mds_request *req;
1927         int err;
1928
1929         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1930                 dout("do_getattr inode %p SNAPDIR\n", inode);
1931                 return 0;
1932         }
1933
1934         dout("do_getattr inode %p mask %s mode 0%o\n",
1935              inode, ceph_cap_string(mask), inode->i_mode);
1936         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1937                 return 0;
1938
1939         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1940         if (IS_ERR(req))
1941                 return PTR_ERR(req);
1942         req->r_inode = inode;
1943         ihold(inode);
1944         req->r_num_caps = 1;
1945         req->r_args.getattr.mask = cpu_to_le32(mask);
1946         req->r_locked_page = locked_page;
1947         err = ceph_mdsc_do_request(mdsc, NULL, req);
1948         if (locked_page && err == 0) {
1949                 u64 inline_version = req->r_reply_info.targeti.inline_version;
1950                 if (inline_version == 0) {
1951                         /* the reply is supposed to contain inline data */
1952                         err = -EINVAL;
1953                 } else if (inline_version == CEPH_INLINE_NONE) {
1954                         err = -ENODATA;
1955                 } else {
1956                         err = req->r_reply_info.targeti.inline_len;
1957                 }
1958         }
1959         ceph_mdsc_put_request(req);
1960         dout("do_getattr result=%d\n", err);
1961         return err;
1962 }
1963
1964
1965 /*
1966  * Check inode permissions.  We verify we have a valid value for
1967  * the AUTH cap, then call the generic handler.
1968  */
1969 int ceph_permission(struct inode *inode, int mask)
1970 {
1971         int err;
1972
1973         if (mask & MAY_NOT_BLOCK)
1974                 return -ECHILD;
1975
1976         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1977
1978         if (!err)
1979                 err = generic_permission(inode, mask);
1980         return err;
1981 }
1982
1983 /*
1984  * Get all attributes.  Hopefully somedata we'll have a statlite()
1985  * and can limit the fields we require to be accurate.
1986  */
1987 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1988                  struct kstat *stat)
1989 {
1990         struct inode *inode = dentry->d_inode;
1991         struct ceph_inode_info *ci = ceph_inode(inode);
1992         int err;
1993
1994         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
1995         if (!err) {
1996                 generic_fillattr(inode, stat);
1997                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1998                 if (ceph_snap(inode) != CEPH_NOSNAP)
1999                         stat->dev = ceph_snap(inode);
2000                 else
2001                         stat->dev = 0;
2002                 if (S_ISDIR(inode->i_mode)) {
2003                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2004                                                 RBYTES))
2005                                 stat->size = ci->i_rbytes;
2006                         else
2007                                 stat->size = ci->i_files + ci->i_subdirs;
2008                         stat->blocks = 0;
2009                         stat->blksize = 65536;
2010                 }
2011         }
2012         return err;
2013 }