fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16
  17 #include "super.h"
  18 #include "mds_client.h"
  19 #include "cache.h"
  20 #include <linux/ceph/decode.h>
  21
  22 /*
  23  * Ceph inode operations
  24  *
  25  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  26  * setattr, etc.), xattr helpers, and helpers for assimilating
  27  * metadata returned by the MDS into our cache.
  28  *
  29  * Also define helpers for doing asynchronous writeback, invalidation,
  30  * and truncation for the benefit of those who can't afford to block
  31  * (typically because they are in the message handler path).
  32  */
  33
  34 static const struct inode_operations ceph_symlink_iops;
  35
  36 static void ceph_invalidate_work(struct work_struct *work);
  37 static void ceph_writeback_work(struct work_struct *work);
  38 static void ceph_vmtruncate_work(struct work_struct *work);
  39
  40 /*
  41  * find or create an inode, given the ceph ino number
  42  */
  43 static int ceph_set_ino_cb(struct inode *inode, void *data)
  44 {
  45         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  46         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  47         return 0;
  48 }
  49
  50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  51 {
  52         struct inode *inode;
  53         ino_t t = ceph_vino_to_ino(vino);
  54
  55         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  56         if (!inode)
  57                 return ERR_PTR(-ENOMEM);
  58         if (inode->i_state & I_NEW) {
  59                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  60                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  61                 unlock_new_inode(inode);
  62         }
  63
  64         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  65              vino.snap, inode);
  66         return inode;
  67 }
  68
  69 /*
  70  * get/constuct snapdir inode for a given directory
  71  */
  72 struct inode *ceph_get_snapdir(struct inode *parent)
  73 {
  74         struct ceph_vino vino = {
  75                 .ino = ceph_ino(parent),
  76                 .snap = CEPH_SNAPDIR,
  77         };
  78         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  79         struct ceph_inode_info *ci = ceph_inode(inode);
  80
  81         BUG_ON(!S_ISDIR(parent->i_mode));
  82         if (IS_ERR(inode))
  83                 return inode;
  84         inode->i_mode = parent->i_mode;
  85         inode->i_uid = parent->i_uid;
  86         inode->i_gid = parent->i_gid;
  87         inode->i_op = &ceph_snapdir_iops;
  88         inode->i_fop = &ceph_snapdir_fops;
  89         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  90         ci->i_rbytes = 0;
  91         return inode;
  92 }
  93
  94 const struct inode_operations ceph_file_iops = {
  95         .permission = ceph_permission,
  96         .setattr = ceph_setattr,
  97         .getattr = ceph_getattr,
  98         .listxattr = ceph_listxattr,
  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                 return ERR_PTR(-ENOMEM);
 139
 140         frag->frag = f;
 141         frag->split_by = 0;
 142         frag->mds = -1;
 143         frag->ndist = 0;
 144
 145         rb_link_node(&frag->node, parent, p);
 146         rb_insert_color(&frag->node, &ci->i_fragtree);
 147
 148         dout("get_or_create_frag added %llx.%llx frag %x\n",
 149              ceph_vinop(&ci->vfs_inode), f);
 150         return frag;
 151 }
 152
 153 /*
 154  * find a specific frag @f
 155  */
 156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 157 {
 158         struct rb_node *n = ci->i_fragtree.rb_node;
 159
 160         while (n) {
 161                 struct ceph_inode_frag *frag =
 162                         rb_entry(n, struct ceph_inode_frag, node);
 163                 int c = ceph_frag_compare(f, frag->frag);
 164                 if (c < 0)
 165                         n = n->rb_left;
 166                 else if (c > 0)
 167                         n = n->rb_right;
 168                 else
 169                         return frag;
 170         }
 171         return NULL;
 172 }
 173
 174 /*
 175  * Choose frag containing the given value @v.  If @pfrag is
 176  * specified, copy the frag delegation info to the caller if
 177  * it is present.
 178  */
 179 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 180                               struct ceph_inode_frag *pfrag, int *found)
 181 {
 182         u32 t = ceph_frag_make(0, 0);
 183         struct ceph_inode_frag *frag;
 184         unsigned nway, i;
 185         u32 n;
 186
 187         if (found)
 188                 *found = 0;
 189
 190         while (1) {
 191                 WARN_ON(!ceph_frag_contains_value(t, v));
 192                 frag = __ceph_find_frag(ci, t);
 193                 if (!frag)
 194                         break; /* t is a leaf */
 195                 if (frag->split_by == 0) {
 196                         if (pfrag)
 197                                 memcpy(pfrag, frag, sizeof(*pfrag));
 198                         if (found)
 199                                 *found = 1;
 200                         break;
 201                 }
 202
 203                 /* choose child */
 204                 nway = 1 << frag->split_by;
 205                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 206                      frag->split_by, nway);
 207                 for (i = 0; i < nway; i++) {
 208                         n = ceph_frag_make_child(t, frag->split_by, i);
 209                         if (ceph_frag_contains_value(n, v)) {
 210                                 t = n;
 211                                 break;
 212                         }
 213                 }
 214                 BUG_ON(i == nway);
 215         }
 216         dout("choose_frag(%x) = %x\n", v, t);
 217
 218         return t;
 219 }
 220
 221 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 222                      struct ceph_inode_frag *pfrag, int *found)
 223 {
 224         u32 ret;
 225         mutex_lock(&ci->i_fragtree_mutex);
 226         ret = __ceph_choose_frag(ci, v, pfrag, found);
 227         mutex_unlock(&ci->i_fragtree_mutex);
 228         return ret;
 229 }
 230
 231 /*
 232  * Process dirfrag (delegation) info from the mds.  Include leaf
 233  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 234  * branches/splits are included in i_fragtree)
 235  */
 236 static int ceph_fill_dirfrag(struct inode *inode,
 237                              struct ceph_mds_reply_dirfrag *dirinfo)
 238 {
 239         struct ceph_inode_info *ci = ceph_inode(inode);
 240         struct ceph_inode_frag *frag;
 241         u32 id = le32_to_cpu(dirinfo->frag);
 242         int mds = le32_to_cpu(dirinfo->auth);
 243         int ndist = le32_to_cpu(dirinfo->ndist);
 244         int diri_auth = -1;
 245         int i;
 246         int err = 0;
 247
 248         spin_lock(&ci->i_ceph_lock);
 249         if (ci->i_auth_cap)
 250                 diri_auth = ci->i_auth_cap->mds;
 251         spin_unlock(&ci->i_ceph_lock);
 252
 253         if (mds == -1) /* CDIR_AUTH_PARENT */
 254                 mds = diri_auth;
 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 frag_tree_split_cmp(const void *l, const void *r)
 303 {
 304         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 305         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 306         return ceph_frag_compare(le32_to_cpu(ls->frag),
 307                                  le32_to_cpu(rs->frag));
 308 }
 309
 310 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 311 {
 312         if (!frag)
 313                 return f == ceph_frag_make(0, 0);
 314         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 315                 return false;
 316         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 317 }
 318
 319 static int ceph_fill_fragtree(struct inode *inode,
 320                               struct ceph_frag_tree_head *fragtree,
 321                               struct ceph_mds_reply_dirfrag *dirinfo)
 322 {
 323         struct ceph_inode_info *ci = ceph_inode(inode);
 324         struct ceph_inode_frag *frag, *prev_frag = NULL;
 325         struct rb_node *rb_node;
 326         unsigned i, split_by, nsplits;
 327         u32 id;
 328         bool update = false;
 329
 330         mutex_lock(&ci->i_fragtree_mutex);
 331         nsplits = le32_to_cpu(fragtree->nsplits);
 332         if (nsplits != ci->i_fragtree_nsplits) {
 333                 update = true;
 334         } else if (nsplits) {
 335                 i = prandom_u32() % nsplits;
 336                 id = le32_to_cpu(fragtree->splits[i].frag);
 337                 if (!__ceph_find_frag(ci, id))
 338                         update = true;
 339         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 340                 rb_node = rb_first(&ci->i_fragtree);
 341                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 342                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 343                         update = true;
 344         }
 345         if (!update && dirinfo) {
 346                 id = le32_to_cpu(dirinfo->frag);
 347                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 348                         update = true;
 349         }
 350         if (!update)
 351                 goto out_unlock;
 352
 353         if (nsplits > 1) {
 354                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 355                      frag_tree_split_cmp, NULL);
 356         }
 357
 358         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 359         rb_node = rb_first(&ci->i_fragtree);
 360         for (i = 0; i < nsplits; i++) {
 361                 id = le32_to_cpu(fragtree->splits[i].frag);
 362                 split_by = le32_to_cpu(fragtree->splits[i].by);
 363                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 364                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 365                                "frag %x split by %d\n", ceph_vinop(inode),
 366                                i, nsplits, id, split_by);
 367                         continue;
 368                 }
 369                 frag = NULL;
 370                 while (rb_node) {
 371                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 372                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 373                                 if (frag->frag != id)
 374                                         frag = NULL;
 375                                 else
 376                                         rb_node = rb_next(rb_node);
 377                                 break;
 378                         }
 379                         rb_node = rb_next(rb_node);
 380                         /* delete stale split/leaf node */
 381                         if (frag->split_by > 0 ||
 382                             !is_frag_child(frag->frag, prev_frag)) {
 383                                 rb_erase(&frag->node, &ci->i_fragtree);
 384                                 if (frag->split_by > 0)
 385                                         ci->i_fragtree_nsplits--;
 386                                 kfree(frag);
 387                         }
 388                         frag = NULL;
 389                 }
 390                 if (!frag) {
 391                         frag = __get_or_create_frag(ci, id);
 392                         if (IS_ERR(frag))
 393                                 continue;
 394                 }
 395                 if (frag->split_by == 0)
 396                         ci->i_fragtree_nsplits++;
 397                 frag->split_by = split_by;
 398                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 399                 prev_frag = frag;
 400         }
 401         while (rb_node) {
 402                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 403                 rb_node = rb_next(rb_node);
 404                 /* delete stale split/leaf node */
 405                 if (frag->split_by > 0 ||
 406                     !is_frag_child(frag->frag, prev_frag)) {
 407                         rb_erase(&frag->node, &ci->i_fragtree);
 408                         if (frag->split_by > 0)
 409                                 ci->i_fragtree_nsplits--;
 410                         kfree(frag);
 411                 }
 412         }
 413 out_unlock:
 414         mutex_unlock(&ci->i_fragtree_mutex);
 415         return 0;
 416 }
 417
 418 /*
 419  * initialize a newly allocated inode.
 420  */
 421 struct inode *ceph_alloc_inode(struct super_block *sb)
 422 {
 423         struct ceph_inode_info *ci;
 424         int i;
 425
 426         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 427         if (!ci)
 428                 return NULL;
 429
 430         dout("alloc_inode %p\n", &ci->vfs_inode);
 431
 432         spin_lock_init(&ci->i_ceph_lock);
 433
 434         ci->i_version = 0;
 435         ci->i_inline_version = 0;
 436         ci->i_time_warp_seq = 0;
 437         ci->i_ceph_flags = 0;
 438         atomic64_set(&ci->i_ordered_count, 1);
 439         atomic64_set(&ci->i_release_count, 1);
 440         atomic64_set(&ci->i_complete_seq[0], 0);
 441         atomic64_set(&ci->i_complete_seq[1], 0);
 442         ci->i_symlink = NULL;
 443
 444         ci->i_max_bytes = 0;
 445         ci->i_max_files = 0;
 446
 447         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 448         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 449
 450         ci->i_fragtree = RB_ROOT;
 451         mutex_init(&ci->i_fragtree_mutex);
 452
 453         ci->i_xattrs.blob = NULL;
 454         ci->i_xattrs.prealloc_blob = NULL;
 455         ci->i_xattrs.dirty = false;
 456         ci->i_xattrs.index = RB_ROOT;
 457         ci->i_xattrs.count = 0;
 458         ci->i_xattrs.names_size = 0;
 459         ci->i_xattrs.vals_size = 0;
 460         ci->i_xattrs.version = 0;
 461         ci->i_xattrs.index_version = 0;
 462
 463         ci->i_caps = RB_ROOT;
 464         ci->i_auth_cap = NULL;
 465         ci->i_dirty_caps = 0;
 466         ci->i_flushing_caps = 0;
 467         INIT_LIST_HEAD(&ci->i_dirty_item);
 468         INIT_LIST_HEAD(&ci->i_flushing_item);
 469         ci->i_prealloc_cap_flush = NULL;
 470         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 471         init_waitqueue_head(&ci->i_cap_wq);
 472         ci->i_hold_caps_min = 0;
 473         ci->i_hold_caps_max = 0;
 474         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 475         INIT_LIST_HEAD(&ci->i_cap_snaps);
 476         ci->i_head_snapc = NULL;
 477         ci->i_snap_caps = 0;
 478
 479         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 480                 ci->i_nr_by_mode[i] = 0;
 481
 482         mutex_init(&ci->i_truncate_mutex);
 483         ci->i_truncate_seq = 0;
 484         ci->i_truncate_size = 0;
 485         ci->i_truncate_pending = 0;
 486
 487         ci->i_max_size = 0;
 488         ci->i_reported_size = 0;
 489         ci->i_wanted_max_size = 0;
 490         ci->i_requested_max_size = 0;
 491
 492         ci->i_pin_ref = 0;
 493         ci->i_rd_ref = 0;
 494         ci->i_rdcache_ref = 0;
 495         ci->i_wr_ref = 0;
 496         ci->i_wb_ref = 0;
 497         ci->i_wrbuffer_ref = 0;
 498         ci->i_wrbuffer_ref_head = 0;
 499         atomic_set(&ci->i_filelock_ref, 0);
 500         atomic_set(&ci->i_shared_gen, 0);
 501         ci->i_rdcache_gen = 0;
 502         ci->i_rdcache_revoking = 0;
 503
 504         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 505         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 506         spin_lock_init(&ci->i_unsafe_lock);
 507
 508         ci->i_snap_realm = NULL;
 509         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 510         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 511
 512         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 513         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 514
 515         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 516
 517         ceph_fscache_inode_init(ci);
 518
 519         return &ci->vfs_inode;
 520 }
 521
 522 static void ceph_i_callback(struct rcu_head *head)
 523 {
 524         struct inode *inode = container_of(head, struct inode, i_rcu);
 525         struct ceph_inode_info *ci = ceph_inode(inode);
 526
 527         kmem_cache_free(ceph_inode_cachep, ci);
 528 }
 529
 530 void ceph_destroy_inode(struct inode *inode)
 531 {
 532         struct ceph_inode_info *ci = ceph_inode(inode);
 533         struct ceph_inode_frag *frag;
 534         struct rb_node *n;
 535
 536         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 537
 538         ceph_fscache_unregister_inode_cookie(ci);
 539
 540         ceph_queue_caps_release(inode);
 541
 542         if (__ceph_has_any_quota(ci))
 543                 ceph_adjust_quota_realms_count(inode, false);
 544
 545         /*
 546          * we may still have a snap_realm reference if there are stray
 547          * caps in i_snap_caps.
 548          */
 549         if (ci->i_snap_realm) {
 550                 struct ceph_mds_client *mdsc =
 551                                         ceph_inode_to_client(inode)->mdsc;
 552                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 553                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 554                         dout(" dropping residual ref to snap realm %p\n",
 555                              realm);
 556                         spin_lock(&realm->inodes_with_caps_lock);
 557                         list_del_init(&ci->i_snap_realm_item);
 558                         ci->i_snap_realm = NULL;
 559                         if (realm->ino == ci->i_vino.ino)
 560                                 realm->inode = NULL;
 561                         spin_unlock(&realm->inodes_with_caps_lock);
 562                         ceph_put_snap_realm(mdsc, realm);
 563                 } else {
 564                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 565                         ci->i_snap_realm = NULL;
 566                 }
 567         }
 568
 569         kfree(ci->i_symlink);
 570         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 571                 frag = rb_entry(n, struct ceph_inode_frag, node);
 572                 rb_erase(n, &ci->i_fragtree);
 573                 kfree(frag);
 574         }
 575         ci->i_fragtree_nsplits = 0;
 576
 577         __ceph_destroy_xattrs(ci);
 578         if (ci->i_xattrs.blob)
 579                 ceph_buffer_put(ci->i_xattrs.blob);
 580         if (ci->i_xattrs.prealloc_blob)
 581                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 582
 583         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 584
 585         call_rcu(&inode->i_rcu, ceph_i_callback);
 586 }
 587
 588 int ceph_drop_inode(struct inode *inode)
 589 {
 590         /*
 591          * Positve dentry and corresponding inode are always accompanied
 592          * in MDS reply. So no need to keep inode in the cache after
 593          * dropping all its aliases.
 594          */
 595         return 1;
 596 }
 597
 598 static inline blkcnt_t calc_inode_blocks(u64 size)
 599 {
 600         return (size + (1<<9) - 1) >> 9;
 601 }
 602
 603 /*
 604  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 605  * careful because either the client or MDS may have more up to date
 606  * info, depending on which capabilities are held, and whether
 607  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 608  * and size are monotonically increasing, except when utimes() or
 609  * truncate() increments the corresponding _seq values.)
 610  */
 611 int ceph_fill_file_size(struct inode *inode, int issued,
 612                         u32 truncate_seq, u64 truncate_size, u64 size)
 613 {
 614         struct ceph_inode_info *ci = ceph_inode(inode);
 615         int queue_trunc = 0;
 616
 617         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 618             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 619                 dout("size %lld -> %llu\n", inode->i_size, size);
 620                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 621                         pr_err("fill_file_size non-zero size for directory\n");
 622                         size = 0;
 623                 }
 624                 i_size_write(inode, size);
 625                 inode->i_blocks = calc_inode_blocks(size);
 626                 ci->i_reported_size = size;
 627                 if (truncate_seq != ci->i_truncate_seq) {
 628                         dout("truncate_seq %u -> %u\n",
 629                              ci->i_truncate_seq, truncate_seq);
 630                         ci->i_truncate_seq = truncate_seq;
 631
 632                         /* the MDS should have revoked these caps */
 633                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 634                                                CEPH_CAP_FILE_RD |
 635                                                CEPH_CAP_FILE_WR |
 636                                                CEPH_CAP_FILE_LAZYIO));
 637                         /*
 638                          * If we hold relevant caps, or in the case where we're
 639                          * not the only client referencing this file and we
 640                          * don't hold those caps, then we need to check whether
 641                          * the file is either opened or mmaped
 642                          */
 643                         if ((issued & (CEPH_CAP_FILE_CACHE|
 644                                        CEPH_CAP_FILE_BUFFER)) ||
 645                             mapping_mapped(inode->i_mapping) ||
 646                             __ceph_caps_file_wanted(ci)) {
 647                                 ci->i_truncate_pending++;
 648                                 queue_trunc = 1;
 649                         }
 650                 }
 651         }
 652         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 653             ci->i_truncate_size != truncate_size) {
 654                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 655                      truncate_size);
 656                 ci->i_truncate_size = truncate_size;
 657         }
 658
 659         if (queue_trunc)
 660                 ceph_fscache_invalidate(inode);
 661
 662         return queue_trunc;
 663 }
 664
 665 void ceph_fill_file_time(struct inode *inode, int issued,
 666                          u64 time_warp_seq, struct timespec64 *ctime,
 667                          struct timespec64 *mtime, struct timespec64 *atime)
 668 {
 669         struct ceph_inode_info *ci = ceph_inode(inode);
 670         int warn = 0;
 671
 672         if (issued & (CEPH_CAP_FILE_EXCL|
 673                       CEPH_CAP_FILE_WR|
 674                       CEPH_CAP_FILE_BUFFER|
 675                       CEPH_CAP_AUTH_EXCL|
 676                       CEPH_CAP_XATTR_EXCL)) {
 677                 if (ci->i_version == 0 ||
 678                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 679                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 680                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 681                              ctime->tv_sec, ctime->tv_nsec);
 682                         inode->i_ctime = *ctime;
 683                 }
 684                 if (ci->i_version == 0 ||
 685                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 686                         /* the MDS did a utimes() */
 687                         dout("mtime %lld.%09ld -> %lld.%09ld "
 688                              "tw %d -> %d\n",
 689                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 690                              mtime->tv_sec, mtime->tv_nsec,
 691                              ci->i_time_warp_seq, (int)time_warp_seq);
 692
 693                         inode->i_mtime = *mtime;
 694                         inode->i_atime = *atime;
 695                         ci->i_time_warp_seq = time_warp_seq;
 696                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 697                         /* nobody did utimes(); take the max */
 698                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 699                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 700                                      inode->i_mtime.tv_sec,
 701                                      inode->i_mtime.tv_nsec,
 702                                      mtime->tv_sec, mtime->tv_nsec);
 703                                 inode->i_mtime = *mtime;
 704                         }
 705                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 706                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 707                                      inode->i_atime.tv_sec,
 708                                      inode->i_atime.tv_nsec,
 709                                      atime->tv_sec, atime->tv_nsec);
 710                                 inode->i_atime = *atime;
 711                         }
 712                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 713                         /* we did a utimes(); ignore mds values */
 714                 } else {
 715                         warn = 1;
 716                 }
 717         } else {
 718                 /* we have no write|excl caps; whatever the MDS says is true */
 719                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 720                         inode->i_ctime = *ctime;
 721                         inode->i_mtime = *mtime;
 722                         inode->i_atime = *atime;
 723                         ci->i_time_warp_seq = time_warp_seq;
 724                 } else {
 725                         warn = 1;
 726                 }
 727         }
 728         if (warn) /* time_warp_seq shouldn't go backwards */
 729                 dout("%p mds time_warp_seq %llu < %u\n",
 730                      inode, time_warp_seq, ci->i_time_warp_seq);
 731 }
 732
 733 /*
 734  * Populate an inode based on info from mds.  May be called on new or
 735  * existing inodes.
 736  */
 737 static int fill_inode(struct inode *inode, struct page *locked_page,
 738                       struct ceph_mds_reply_info_in *iinfo,
 739                       struct ceph_mds_reply_dirfrag *dirinfo,
 740                       struct ceph_mds_session *session,
 741                       unsigned long ttl_from, int cap_fmode,
 742                       struct ceph_cap_reservation *caps_reservation)
 743 {
 744         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 745         struct ceph_mds_reply_inode *info = iinfo->in;
 746         struct ceph_inode_info *ci = ceph_inode(inode);
 747         int issued, new_issued, info_caps;
 748         struct timespec64 mtime, atime, ctime;
 749         struct ceph_buffer *xattr_blob = NULL;
 750         struct ceph_string *pool_ns = NULL;
 751         struct ceph_cap *new_cap = NULL;
 752         int err = 0;
 753         bool wake = false;
 754         bool queue_trunc = false;
 755         bool new_version = false;
 756         bool fill_inline = false;
 757
 758         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 759              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 760              ci->i_version);
 761
 762         info_caps = le32_to_cpu(info->cap.caps);
 763
 764         /* prealloc new cap struct */
 765         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
 766                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 767
 768         /*
 769          * prealloc xattr data, if it looks like we'll need it.  only
 770          * if len > 4 (meaning there are actually xattrs; the first 4
 771          * bytes are the xattr count).
 772          */
 773         if (iinfo->xattr_len > 4) {
 774                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 775                 if (!xattr_blob)
 776                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 777                                iinfo->xattr_len);
 778         }
 779
 780         if (iinfo->pool_ns_len > 0)
 781                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 782                                                      iinfo->pool_ns_len);
 783
 784         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 785                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 786
 787         spin_lock(&ci->i_ceph_lock);
 788
 789         /*
 790          * provided version will be odd if inode value is projected,
 791          * even if stable.  skip the update if we have newer stable
 792          * info (ours>=theirs, e.g. due to racing mds replies), unless
 793          * we are getting projected (unstable) info (in which case the
 794          * version is odd, and we want ours>theirs).
 795          *   us   them
 796          *   2    2     skip
 797          *   3    2     skip
 798          *   3    3     update
 799          */
 800         if (ci->i_version == 0 ||
 801             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 802              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 803                 new_version = true;
 804
 805         __ceph_caps_issued(ci, &issued);
 806         issued |= __ceph_caps_dirty(ci);
 807         new_issued = ~issued & info_caps;
 808
 809         /* update inode */
 810         inode->i_rdev = le32_to_cpu(info->rdev);
 811         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 812
 813         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 814
 815         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 816             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 817                 inode->i_mode = le32_to_cpu(info->mode);
 818                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 819                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 820                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 821                      from_kuid(&init_user_ns, inode->i_uid),
 822                      from_kgid(&init_user_ns, inode->i_gid));
 823         }
 824
 825         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 826             (issued & CEPH_CAP_LINK_EXCL) == 0)
 827                 set_nlink(inode, le32_to_cpu(info->nlink));
 828
 829         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 830                 /* be careful with mtime, atime, size */
 831                 ceph_decode_timespec64(&atime, &info->atime);
 832                 ceph_decode_timespec64(&mtime, &info->mtime);
 833                 ceph_decode_timespec64(&ctime, &info->ctime);
 834                 ceph_fill_file_time(inode, issued,
 835                                 le32_to_cpu(info->time_warp_seq),
 836                                 &ctime, &mtime, &atime);
 837         }
 838
 839         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 840                 ci->i_files = le64_to_cpu(info->files);
 841                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 842         }
 843
 844         if (new_version ||
 845             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 846                 s64 old_pool = ci->i_layout.pool_id;
 847                 struct ceph_string *old_ns;
 848
 849                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 850                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 851                                         lockdep_is_held(&ci->i_ceph_lock));
 852                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 853
 854                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 855                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 856
 857                 pool_ns = old_ns;
 858
 859                 queue_trunc = ceph_fill_file_size(inode, issued,
 860                                         le32_to_cpu(info->truncate_seq),
 861                                         le64_to_cpu(info->truncate_size),
 862                                         le64_to_cpu(info->size));
 863                 /* only update max_size on auth cap */
 864                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 865                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 866                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 867                                         le64_to_cpu(info->max_size));
 868                         ci->i_max_size = le64_to_cpu(info->max_size);
 869                 }
 870         }
 871
 872         /* layout and rstat are not tracked by capability, update them if
 873          * the inode info is from auth mds */
 874         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 875                 if (S_ISDIR(inode->i_mode)) {
 876                         ci->i_dir_layout = iinfo->dir_layout;
 877                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 878                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 879                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 880                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 881                 }
 882         }
 883
 884         /* xattrs */
 885         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 886         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 887             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 888                 if (ci->i_xattrs.blob)
 889                         ceph_buffer_put(ci->i_xattrs.blob);
 890                 ci->i_xattrs.blob = xattr_blob;
 891                 if (xattr_blob)
 892                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 893                                iinfo->xattr_data, iinfo->xattr_len);
 894                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 895                 ceph_forget_all_cached_acls(inode);
 896                 xattr_blob = NULL;
 897         }
 898
 899         /* finally update i_version */
 900         if (le64_to_cpu(info->version) > ci->i_version)
 901                 ci->i_version = le64_to_cpu(info->version);
 902
 903         inode->i_mapping->a_ops = &ceph_aops;
 904
 905         switch (inode->i_mode & S_IFMT) {
 906         case S_IFIFO:
 907         case S_IFBLK:
 908         case S_IFCHR:
 909         case S_IFSOCK:
 910                 inode->i_blkbits = PAGE_SHIFT;
 911                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 912                 inode->i_op = &ceph_file_iops;
 913                 break;
 914         case S_IFREG:
 915                 inode->i_op = &ceph_file_iops;
 916                 inode->i_fop = &ceph_file_fops;
 917                 break;
 918         case S_IFLNK:
 919                 inode->i_op = &ceph_symlink_iops;
 920                 if (!ci->i_symlink) {
 921                         u32 symlen = iinfo->symlink_len;
 922                         char *sym;
 923
 924                         spin_unlock(&ci->i_ceph_lock);
 925
 926                         if (symlen != i_size_read(inode)) {
 927                                 pr_err("fill_inode %llx.%llx BAD symlink "
 928                                         "size %lld\n", ceph_vinop(inode),
 929                                         i_size_read(inode));
 930                                 i_size_write(inode, symlen);
 931                                 inode->i_blocks = calc_inode_blocks(symlen);
 932                         }
 933
 934                         err = -ENOMEM;
 935                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 936                         if (!sym)
 937                                 goto out;
 938
 939                         spin_lock(&ci->i_ceph_lock);
 940                         if (!ci->i_symlink)
 941                                 ci->i_symlink = sym;
 942                         else
 943                                 kfree(sym); /* lost a race */
 944                 }
 945                 inode->i_link = ci->i_symlink;
 946                 break;
 947         case S_IFDIR:
 948                 inode->i_op = &ceph_dir_iops;
 949                 inode->i_fop = &ceph_dir_fops;
 950                 break;
 951         default:
 952                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 953                        ceph_vinop(inode), inode->i_mode);
 954         }
 955
 956         /* were we issued a capability? */
 957         if (info_caps) {
 958                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 959                         ceph_add_cap(inode, session,
 960                                      le64_to_cpu(info->cap.cap_id),
 961                                      cap_fmode, info_caps,
 962                                      le32_to_cpu(info->cap.wanted),
 963                                      le32_to_cpu(info->cap.seq),
 964                                      le32_to_cpu(info->cap.mseq),
 965                                      le64_to_cpu(info->cap.realm),
 966                                      info->cap.flags, &new_cap);
 967
 968                         /* set dir completion flag? */
 969                         if (S_ISDIR(inode->i_mode) &&
 970                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 971                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 972                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 973                             !__ceph_dir_is_complete(ci)) {
 974                                 dout(" marking %p complete (empty)\n", inode);
 975                                 i_size_write(inode, 0);
 976                                 __ceph_dir_set_complete(ci,
 977                                         atomic64_read(&ci->i_release_count),
 978                                         atomic64_read(&ci->i_ordered_count));
 979                         }
 980
 981                         wake = true;
 982                 } else {
 983                         dout(" %p got snap_caps %s\n", inode,
 984                              ceph_cap_string(info_caps));
 985                         ci->i_snap_caps |= info_caps;
 986                         if (cap_fmode >= 0)
 987                                 __ceph_get_fmode(ci, cap_fmode);
 988                 }
 989         } else if (cap_fmode >= 0) {
 990                 pr_warn("mds issued no caps on %llx.%llx\n",
 991                            ceph_vinop(inode));
 992                 __ceph_get_fmode(ci, cap_fmode);
 993         }
 994
 995         if (iinfo->inline_version > 0 &&
 996             iinfo->inline_version >= ci->i_inline_version) {
 997                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 998                 ci->i_inline_version = iinfo->inline_version;
 999                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1000                     (locked_page || (info_caps & cache_caps)))
1001                         fill_inline = true;
1002         }
1003
1004         spin_unlock(&ci->i_ceph_lock);
1005
1006         if (fill_inline)
1007                 ceph_fill_inline_data(inode, locked_page,
1008                                       iinfo->inline_data, iinfo->inline_len);
1009
1010         if (wake)
1011                 wake_up_all(&ci->i_cap_wq);
1012
1013         /* queue truncate if we saw i_size decrease */
1014         if (queue_trunc)
1015                 ceph_queue_vmtruncate(inode);
1016
1017         /* populate frag tree */
1018         if (S_ISDIR(inode->i_mode))
1019                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1020
1021         /* update delegation info? */
1022         if (dirinfo)
1023                 ceph_fill_dirfrag(inode, dirinfo);
1024
1025         err = 0;
1026 out:
1027         if (new_cap)
1028                 ceph_put_cap(mdsc, new_cap);
1029         if (xattr_blob)
1030                 ceph_buffer_put(xattr_blob);
1031         ceph_put_string(pool_ns);
1032         return err;
1033 }
1034
1035 /*
1036  * caller should hold session s_mutex.
1037  */
1038 static void update_dentry_lease(struct dentry *dentry,
1039                                 struct ceph_mds_reply_lease *lease,
1040                                 struct ceph_mds_session *session,
1041                                 unsigned long from_time,
1042                                 struct ceph_vino *tgt_vino,
1043                                 struct ceph_vino *dir_vino)
1044 {
1045         struct ceph_dentry_info *di = ceph_dentry(dentry);
1046         long unsigned duration = le32_to_cpu(lease->duration_ms);
1047         long unsigned ttl = from_time + (duration * HZ) / 1000;
1048         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1049         struct inode *dir;
1050         struct ceph_mds_session *old_lease_session = NULL;
1051
1052         /*
1053          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1054          * we expect a negative dentry.
1055          */
1056         if (!tgt_vino && d_really_is_positive(dentry))
1057                 return;
1058
1059         if (tgt_vino && (d_really_is_negative(dentry) ||
1060                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1061                 return;
1062
1063         spin_lock(&dentry->d_lock);
1064         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1065              dentry, duration, ttl);
1066
1067         dir = d_inode(dentry->d_parent);
1068
1069         /* make sure parent matches dir_vino */
1070         if (!ceph_ino_compare(dir, dir_vino))
1071                 goto out_unlock;
1072
1073         /* only track leases on regular dentries */
1074         if (ceph_snap(dir) != CEPH_NOSNAP)
1075                 goto out_unlock;
1076
1077         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1078
1079         if (duration == 0)
1080                 goto out_unlock;
1081
1082         if (di->lease_gen == session->s_cap_gen &&
1083             time_before(ttl, di->time))
1084                 goto out_unlock;  /* we already have a newer lease. */
1085
1086         if (di->lease_session && di->lease_session != session) {
1087                 old_lease_session = di->lease_session;
1088                 di->lease_session = NULL;
1089         }
1090
1091         ceph_dentry_lru_touch(dentry);
1092
1093         if (!di->lease_session)
1094                 di->lease_session = ceph_get_mds_session(session);
1095         di->lease_gen = session->s_cap_gen;
1096         di->lease_seq = le32_to_cpu(lease->seq);
1097         di->lease_renew_after = half_ttl;
1098         di->lease_renew_from = 0;
1099         di->time = ttl;
1100 out_unlock:
1101         spin_unlock(&dentry->d_lock);
1102         if (old_lease_session)
1103                 ceph_put_mds_session(old_lease_session);
1104 }
1105
1106 /*
1107  * splice a dentry to an inode.
1108  * caller must hold directory i_mutex for this to be safe.
1109  */
1110 static int splice_dentry(struct dentry **pdn, struct inode *in)
1111 {
1112         struct dentry *dn = *pdn;
1113         struct dentry *realdn;
1114
1115         BUG_ON(d_inode(dn));
1116
1117         if (S_ISDIR(in->i_mode)) {
1118                 /* If inode is directory, d_splice_alias() below will remove
1119                  * 'realdn' from its origin parent. We need to ensure that
1120                  * origin parent's readdir cache will not reference 'realdn'
1121                  */
1122                 realdn = d_find_any_alias(in);
1123                 if (realdn) {
1124                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1125                         spin_lock(&realdn->d_lock);
1126
1127                         realdn->d_op->d_prune(realdn);
1128
1129                         di->time = jiffies;
1130                         di->lease_shared_gen = 0;
1131                         di->offset = 0;
1132
1133                         spin_unlock(&realdn->d_lock);
1134                         dput(realdn);
1135                 }
1136         }
1137
1138         /* dn must be unhashed */
1139         if (!d_unhashed(dn))
1140                 d_drop(dn);
1141         realdn = d_splice_alias(in, dn);
1142         if (IS_ERR(realdn)) {
1143                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1144                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1145                 return PTR_ERR(realdn);
1146         }
1147
1148         if (realdn) {
1149                 dout("dn %p (%d) spliced with %p (%d) "
1150                      "inode %p ino %llx.%llx\n",
1151                      dn, d_count(dn),
1152                      realdn, d_count(realdn),
1153                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1154                 dput(dn);
1155                 *pdn = realdn;
1156         } else {
1157                 BUG_ON(!ceph_dentry(dn));
1158                 dout("dn %p attached to %p ino %llx.%llx\n",
1159                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1160         }
1161         return 0;
1162 }
1163
1164 /*
1165  * Incorporate results into the local cache.  This is either just
1166  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1167  * after a lookup).
1168  *
1169  * A reply may contain
1170  *         a directory inode along with a dentry.
1171  *  and/or a target inode
1172  *
1173  * Called with snap_rwsem (read).
1174  */
1175 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1176 {
1177         struct ceph_mds_session *session = req->r_session;
1178         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1179         struct inode *in = NULL;
1180         struct ceph_vino tvino, dvino;
1181         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1182         int err = 0;
1183
1184         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1185              rinfo->head->is_dentry, rinfo->head->is_target);
1186
1187         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1188                 dout("fill_trace reply is empty!\n");
1189                 if (rinfo->head->result == 0 && req->r_parent)
1190                         ceph_invalidate_dir_request(req);
1191                 return 0;
1192         }
1193
1194         if (rinfo->head->is_dentry) {
1195                 struct inode *dir = req->r_parent;
1196
1197                 if (dir) {
1198                         err = fill_inode(dir, NULL,
1199                                          &rinfo->diri, rinfo->dirfrag,
1200                                          session, req->r_request_started, -1,
1201                                          &req->r_caps_reservation);
1202                         if (err < 0)
1203                                 goto done;
1204                 } else {
1205                         WARN_ON_ONCE(1);
1206                 }
1207
1208                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1209                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1210                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1211                         struct qstr dname;
1212                         struct dentry *dn, *parent;
1213
1214                         BUG_ON(!rinfo->head->is_target);
1215                         BUG_ON(req->r_dentry);
1216
1217                         parent = d_find_any_alias(dir);
1218                         BUG_ON(!parent);
1219
1220                         dname.name = rinfo->dname;
1221                         dname.len = rinfo->dname_len;
1222                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1223                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1224                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1225 retry_lookup:
1226                         dn = d_lookup(parent, &dname);
1227                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1228                              parent, dname.len, dname.name, dn);
1229
1230                         if (!dn) {
1231                                 dn = d_alloc(parent, &dname);
1232                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1233                                      dname.len, dname.name, dn);
1234                                 if (!dn) {
1235                                         dput(parent);
1236                                         err = -ENOMEM;
1237                                         goto done;
1238                                 }
1239                                 err = 0;
1240                         } else if (d_really_is_positive(dn) &&
1241                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1242                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1243                                 dout(" dn %p points to wrong inode %p\n",
1244                                      dn, d_inode(dn));
1245                                 ceph_dir_clear_ordered(dir);
1246                                 d_delete(dn);
1247                                 dput(dn);
1248                                 goto retry_lookup;
1249                         }
1250
1251                         req->r_dentry = dn;
1252                         dput(parent);
1253                 }
1254         }
1255
1256         if (rinfo->head->is_target) {
1257                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1258                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1259
1260                 in = ceph_get_inode(sb, tvino);
1261                 if (IS_ERR(in)) {
1262                         err = PTR_ERR(in);
1263                         goto done;
1264                 }
1265                 req->r_target_inode = in;
1266
1267                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1268                                 session, req->r_request_started,
1269                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1270                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1271                                 &req->r_caps_reservation);
1272                 if (err < 0) {
1273                         pr_err("fill_inode badness %p %llx.%llx\n",
1274                                 in, ceph_vinop(in));
1275                         goto done;
1276                 }
1277         }
1278
1279         /*
1280          * ignore null lease/binding on snapdir ENOENT, or else we
1281          * will have trouble splicing in the virtual snapdir later
1282          */
1283         if (rinfo->head->is_dentry &&
1284             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1285             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1286             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1287                                                fsc->mount_options->snapdir_name,
1288                                                req->r_dentry->d_name.len))) {
1289                 /*
1290                  * lookup link rename   : null -> possibly existing inode
1291                  * mknod symlink mkdir  : null -> new inode
1292                  * unlink               : linked -> null
1293                  */
1294                 struct inode *dir = req->r_parent;
1295                 struct dentry *dn = req->r_dentry;
1296                 bool have_dir_cap, have_lease;
1297
1298                 BUG_ON(!dn);
1299                 BUG_ON(!dir);
1300                 BUG_ON(d_inode(dn->d_parent) != dir);
1301
1302                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1303                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1304
1305                 BUG_ON(ceph_ino(dir) != dvino.ino);
1306                 BUG_ON(ceph_snap(dir) != dvino.snap);
1307
1308                 /* do we have a lease on the whole dir? */
1309                 have_dir_cap =
1310                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1311                          CEPH_CAP_FILE_SHARED);
1312
1313                 /* do we have a dn lease? */
1314                 have_lease = have_dir_cap ||
1315                         le32_to_cpu(rinfo->dlease->duration_ms);
1316                 if (!have_lease)
1317                         dout("fill_trace  no dentry lease or dir cap\n");
1318
1319                 /* rename? */
1320                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1321                         struct inode *olddir = req->r_old_dentry_dir;
1322                         BUG_ON(!olddir);
1323
1324                         dout(" src %p '%pd' dst %p '%pd'\n",
1325                              req->r_old_dentry,
1326                              req->r_old_dentry,
1327                              dn, dn);
1328                         dout("fill_trace doing d_move %p -> %p\n",
1329                              req->r_old_dentry, dn);
1330
1331                         /* d_move screws up sibling dentries' offsets */
1332                         ceph_dir_clear_ordered(dir);
1333                         ceph_dir_clear_ordered(olddir);
1334
1335                         d_move(req->r_old_dentry, dn);
1336                         dout(" src %p '%pd' dst %p '%pd'\n",
1337                              req->r_old_dentry,
1338                              req->r_old_dentry,
1339                              dn, dn);
1340
1341                         /* ensure target dentry is invalidated, despite
1342                            rehashing bug in vfs_rename_dir */
1343                         ceph_invalidate_dentry_lease(dn);
1344
1345                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1346                              ceph_dentry(req->r_old_dentry)->offset);
1347
1348                         /* swap r_dentry and r_old_dentry in case that
1349                          * splice_dentry() gets called later. This is safe
1350                          * because no other place will use them */
1351                         req->r_dentry = req->r_old_dentry;
1352                         req->r_old_dentry = dn;
1353                         dn = req->r_dentry;
1354                 }
1355
1356                 /* null dentry? */
1357                 if (!rinfo->head->is_target) {
1358                         dout("fill_trace null dentry\n");
1359                         if (d_really_is_positive(dn)) {
1360                                 dout("d_delete %p\n", dn);
1361                                 ceph_dir_clear_ordered(dir);
1362                                 d_delete(dn);
1363                         } else if (have_lease) {
1364                                 if (d_unhashed(dn))
1365                                         d_add(dn, NULL);
1366                                 update_dentry_lease(dn, rinfo->dlease,
1367                                                     session,
1368                                                     req->r_request_started,
1369                                                     NULL, &dvino);
1370                         }
1371                         goto done;
1372                 }
1373
1374                 /* attach proper inode */
1375                 if (d_really_is_negative(dn)) {
1376                         ceph_dir_clear_ordered(dir);
1377                         ihold(in);
1378                         err = splice_dentry(&req->r_dentry, in);
1379                         if (err < 0)
1380                                 goto done;
1381                         dn = req->r_dentry;  /* may have spliced */
1382                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1383                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1384                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1385                              ceph_vinop(in));
1386                         d_invalidate(dn);
1387                         have_lease = false;
1388                 }
1389
1390                 if (have_lease) {
1391                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1392                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1393                         update_dentry_lease(dn, rinfo->dlease, session,
1394                                             req->r_request_started,
1395                                             &tvino, &dvino);
1396                 }
1397                 dout(" final dn %p\n", dn);
1398         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1399                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1400                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1401                 struct inode *dir = req->r_parent;
1402
1403                 /* fill out a snapdir LOOKUPSNAP dentry */
1404                 BUG_ON(!dir);
1405                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1406                 BUG_ON(!req->r_dentry);
1407                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1408                 ceph_dir_clear_ordered(dir);
1409                 ihold(in);
1410                 err = splice_dentry(&req->r_dentry, in);
1411                 if (err < 0)
1412                         goto done;
1413         } else if (rinfo->head->is_dentry) {
1414                 struct ceph_vino *ptvino = NULL;
1415
1416                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1417                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1418                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1419                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1420
1421                         if (rinfo->head->is_target) {
1422                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1423                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1424                                 ptvino = &tvino;
1425                         }
1426
1427                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1428                                 session, req->r_request_started, ptvino,
1429                                 &dvino);
1430                 } else {
1431                         dout("%s: no dentry lease or dir cap\n", __func__);
1432                 }
1433         }
1434 done:
1435         dout("fill_trace done err=%d\n", err);
1436         return err;
1437 }
1438
1439 /*
1440  * Prepopulate our cache with readdir results, leases, etc.
1441  */
1442 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1443                                            struct ceph_mds_session *session)
1444 {
1445         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1446         int i, err = 0;
1447
1448         for (i = 0; i < rinfo->dir_nr; i++) {
1449                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1450                 struct ceph_vino vino;
1451                 struct inode *in;
1452                 int rc;
1453
1454                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1455                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1456
1457                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1458                 if (IS_ERR(in)) {
1459                         err = PTR_ERR(in);
1460                         dout("new_inode badness got %d\n", err);
1461                         continue;
1462                 }
1463                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1464                                 req->r_request_started, -1,
1465                                 &req->r_caps_reservation);
1466                 if (rc < 0) {
1467                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1468                         err = rc;
1469                 }
1470                 iput(in);
1471         }
1472
1473         return err;
1474 }
1475
1476 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1477 {
1478         if (ctl->page) {
1479                 kunmap(ctl->page);
1480                 put_page(ctl->page);
1481                 ctl->page = NULL;
1482         }
1483 }
1484
1485 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1486                               struct ceph_readdir_cache_control *ctl,
1487                               struct ceph_mds_request *req)
1488 {
1489         struct ceph_inode_info *ci = ceph_inode(dir);
1490         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1491         unsigned idx = ctl->index % nsize;
1492         pgoff_t pgoff = ctl->index / nsize;
1493
1494         if (!ctl->page || pgoff != page_index(ctl->page)) {
1495                 ceph_readdir_cache_release(ctl);
1496                 if (idx == 0)
1497                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1498                 else
1499                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1500                 if (!ctl->page) {
1501                         ctl->index = -1;
1502                         return idx == 0 ? -ENOMEM : 0;
1503                 }
1504                 /* reading/filling the cache are serialized by
1505                  * i_mutex, no need to use page lock */
1506                 unlock_page(ctl->page);
1507                 ctl->dentries = kmap(ctl->page);
1508                 if (idx == 0)
1509                         memset(ctl->dentries, 0, PAGE_SIZE);
1510         }
1511
1512         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1513             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1514                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1515                 ctl->dentries[idx] = dn;
1516                 ctl->index++;
1517         } else {
1518                 dout("disable readdir cache\n");
1519                 ctl->index = -1;
1520         }
1521         return 0;
1522 }
1523
1524 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1525                              struct ceph_mds_session *session)
1526 {
1527         struct dentry *parent = req->r_dentry;
1528         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1529         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1530         struct qstr dname;
1531         struct dentry *dn;
1532         struct inode *in;
1533         int err = 0, skipped = 0, ret, i;
1534         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1535         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1536         u32 last_hash = 0;
1537         u32 fpos_offset;
1538         struct ceph_readdir_cache_control cache_ctl = {};
1539
1540         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1541                 return readdir_prepopulate_inodes_only(req, session);
1542
1543         if (rinfo->hash_order) {
1544                 if (req->r_path2) {
1545                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1546                                                   req->r_path2,
1547                                                   strlen(req->r_path2));
1548                         last_hash = ceph_frag_value(last_hash);
1549                 } else if (rinfo->offset_hash) {
1550                         /* mds understands offset_hash */
1551                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1552                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1553                 }
1554         }
1555
1556         if (rinfo->dir_dir &&
1557             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1558                 dout("readdir_prepopulate got new frag %x -> %x\n",
1559                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1560                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1561                 if (!rinfo->hash_order)
1562                         req->r_readdir_offset = 2;
1563         }
1564
1565         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1566                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1567                      rinfo->dir_nr, parent);
1568         } else {
1569                 dout("readdir_prepopulate %d items under dn %p\n",
1570                      rinfo->dir_nr, parent);
1571                 if (rinfo->dir_dir)
1572                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1573
1574                 if (ceph_frag_is_leftmost(frag) &&
1575                     req->r_readdir_offset == 2 &&
1576                     !(rinfo->hash_order && last_hash)) {
1577                         /* note dir version at start of readdir so we can
1578                          * tell if any dentries get dropped */
1579                         req->r_dir_release_cnt =
1580                                 atomic64_read(&ci->i_release_count);
1581                         req->r_dir_ordered_cnt =
1582                                 atomic64_read(&ci->i_ordered_count);
1583                         req->r_readdir_cache_idx = 0;
1584                 }
1585         }
1586
1587         cache_ctl.index = req->r_readdir_cache_idx;
1588         fpos_offset = req->r_readdir_offset;
1589
1590         /* FIXME: release caps/leases if error occurs */
1591         for (i = 0; i < rinfo->dir_nr; i++) {
1592                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1593                 struct ceph_vino tvino, dvino;
1594
1595                 dname.name = rde->name;
1596                 dname.len = rde->name_len;
1597                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1598
1599                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1600                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1601
1602                 if (rinfo->hash_order) {
1603                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1604                                                  rde->name, rde->name_len);
1605                         hash = ceph_frag_value(hash);
1606                         if (hash != last_hash)
1607                                 fpos_offset = 2;
1608                         last_hash = hash;
1609                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1610                 } else {
1611                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1612                 }
1613
1614 retry_lookup:
1615                 dn = d_lookup(parent, &dname);
1616                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1617                      parent, dname.len, dname.name, dn);
1618
1619                 if (!dn) {
1620                         dn = d_alloc(parent, &dname);
1621                         dout("d_alloc %p '%.*s' = %p\n", parent,
1622                              dname.len, dname.name, dn);
1623                         if (!dn) {
1624                                 dout("d_alloc badness\n");
1625                                 err = -ENOMEM;
1626                                 goto out;
1627                         }
1628                 } else if (d_really_is_positive(dn) &&
1629                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1630                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1631                         struct ceph_dentry_info *di = ceph_dentry(dn);
1632                         dout(" dn %p points to wrong inode %p\n",
1633                              dn, d_inode(dn));
1634
1635                         spin_lock(&dn->d_lock);
1636                         if (di->offset > 0 &&
1637                             di->lease_shared_gen ==
1638                             atomic_read(&ci->i_shared_gen)) {
1639                                 __ceph_dir_clear_ordered(ci);
1640                                 di->offset = 0;
1641                         }
1642                         spin_unlock(&dn->d_lock);
1643
1644                         d_delete(dn);
1645                         dput(dn);
1646                         goto retry_lookup;
1647                 }
1648
1649                 /* inode */
1650                 if (d_really_is_positive(dn)) {
1651                         in = d_inode(dn);
1652                 } else {
1653                         in = ceph_get_inode(parent->d_sb, tvino);
1654                         if (IS_ERR(in)) {
1655                                 dout("new_inode badness\n");
1656                                 d_drop(dn);
1657                                 dput(dn);
1658                                 err = PTR_ERR(in);
1659                                 goto out;
1660                         }
1661                 }
1662
1663                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1664                                  req->r_request_started, -1,
1665                                  &req->r_caps_reservation);
1666                 if (ret < 0) {
1667                         pr_err("fill_inode badness on %p\n", in);
1668                         if (d_really_is_negative(dn))
1669                                 iput(in);
1670                         d_drop(dn);
1671                         err = ret;
1672                         goto next_item;
1673                 }
1674
1675                 if (d_really_is_negative(dn)) {
1676                         if (ceph_security_xattr_deadlock(in)) {
1677                                 dout(" skip splicing dn %p to inode %p"
1678                                      " (security xattr deadlock)\n", dn, in);
1679                                 iput(in);
1680                                 skipped++;
1681                                 goto next_item;
1682                         }
1683
1684                         err = splice_dentry(&dn, in);
1685                         if (err < 0)
1686                                 goto next_item;
1687                 }
1688
1689                 ceph_dentry(dn)->offset = rde->offset;
1690
1691                 dvino = ceph_vino(d_inode(parent));
1692                 update_dentry_lease(dn, rde->lease, req->r_session,
1693                                     req->r_request_started, &tvino, &dvino);
1694
1695                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1696                         ret = fill_readdir_cache(d_inode(parent), dn,
1697                                                  &cache_ctl, req);
1698                         if (ret < 0)
1699                                 err = ret;
1700                 }
1701 next_item:
1702                 dput(dn);
1703         }
1704 out:
1705         if (err == 0 && skipped == 0) {
1706                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1707                 req->r_readdir_cache_idx = cache_ctl.index;
1708         }
1709         ceph_readdir_cache_release(&cache_ctl);
1710         dout("readdir_prepopulate done\n");
1711         return err;
1712 }
1713
1714 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1715 {
1716         struct ceph_inode_info *ci = ceph_inode(inode);
1717         bool ret;
1718
1719         spin_lock(&ci->i_ceph_lock);
1720         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1721         i_size_write(inode, size);
1722         inode->i_blocks = calc_inode_blocks(size);
1723
1724         ret = __ceph_should_report_size(ci);
1725
1726         spin_unlock(&ci->i_ceph_lock);
1727         return ret;
1728 }
1729
1730 /*
1731  * Write back inode data in a worker thread.  (This can't be done
1732  * in the message handler context.)
1733  */
1734 void ceph_queue_writeback(struct inode *inode)
1735 {
1736         ihold(inode);
1737         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1738                        &ceph_inode(inode)->i_wb_work)) {
1739                 dout("ceph_queue_writeback %p\n", inode);
1740         } else {
1741                 dout("ceph_queue_writeback %p failed\n", inode);
1742                 iput(inode);
1743         }
1744 }
1745
1746 static void ceph_writeback_work(struct work_struct *work)
1747 {
1748         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1749                                                   i_wb_work);
1750         struct inode *inode = &ci->vfs_inode;
1751
1752         dout("writeback %p\n", inode);
1753         filemap_fdatawrite(&inode->i_data);
1754         iput(inode);
1755 }
1756
1757 /*
1758  * queue an async invalidation
1759  */
1760 void ceph_queue_invalidate(struct inode *inode)
1761 {
1762         ihold(inode);
1763         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1764                        &ceph_inode(inode)->i_pg_inv_work)) {
1765                 dout("ceph_queue_invalidate %p\n", inode);
1766         } else {
1767                 dout("ceph_queue_invalidate %p failed\n", inode);
1768                 iput(inode);
1769         }
1770 }
1771
1772 /*
1773  * Invalidate inode pages in a worker thread.  (This can't be done
1774  * in the message handler context.)
1775  */
1776 static void ceph_invalidate_work(struct work_struct *work)
1777 {
1778         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1779                                                   i_pg_inv_work);
1780         struct inode *inode = &ci->vfs_inode;
1781         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1782         u32 orig_gen;
1783         int check = 0;
1784
1785         mutex_lock(&ci->i_truncate_mutex);
1786
1787         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1788                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1789                                     inode, ceph_ino(inode));
1790                 mapping_set_error(inode->i_mapping, -EIO);
1791                 truncate_pagecache(inode, 0);
1792                 mutex_unlock(&ci->i_truncate_mutex);
1793                 goto out;
1794         }
1795
1796         spin_lock(&ci->i_ceph_lock);
1797         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1798              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1799         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1800                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1801                         check = 1;
1802                 spin_unlock(&ci->i_ceph_lock);
1803                 mutex_unlock(&ci->i_truncate_mutex);
1804                 goto out;
1805         }
1806         orig_gen = ci->i_rdcache_gen;
1807         spin_unlock(&ci->i_ceph_lock);
1808
1809         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1810                 pr_err("invalidate_pages %p fails\n", inode);
1811         }
1812
1813         spin_lock(&ci->i_ceph_lock);
1814         if (orig_gen == ci->i_rdcache_gen &&
1815             orig_gen == ci->i_rdcache_revoking) {
1816                 dout("invalidate_pages %p gen %d successful\n", inode,
1817                      ci->i_rdcache_gen);
1818                 ci->i_rdcache_revoking--;
1819                 check = 1;
1820         } else {
1821                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1822                      inode, orig_gen, ci->i_rdcache_gen,
1823                      ci->i_rdcache_revoking);
1824                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1825                         check = 1;
1826         }
1827         spin_unlock(&ci->i_ceph_lock);
1828         mutex_unlock(&ci->i_truncate_mutex);
1829 out:
1830         if (check)
1831                 ceph_check_caps(ci, 0, NULL);
1832         iput(inode);
1833 }
1834
1835
1836 /*
1837  * called by trunc_wq;
1838  *
1839  * We also truncate in a separate thread as well.
1840  */
1841 static void ceph_vmtruncate_work(struct work_struct *work)
1842 {
1843         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1844                                                   i_vmtruncate_work);
1845         struct inode *inode = &ci->vfs_inode;
1846
1847         dout("vmtruncate_work %p\n", inode);
1848         __ceph_do_pending_vmtruncate(inode);
1849         iput(inode);
1850 }
1851
1852 /*
1853  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1854  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1855  */
1856 void ceph_queue_vmtruncate(struct inode *inode)
1857 {
1858         struct ceph_inode_info *ci = ceph_inode(inode);
1859
1860         ihold(inode);
1861
1862         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1863                        &ci->i_vmtruncate_work)) {
1864                 dout("ceph_queue_vmtruncate %p\n", inode);
1865         } else {
1866                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1867                      inode, ci->i_truncate_pending);
1868                 iput(inode);
1869         }
1870 }
1871
1872 /*
1873  * Make sure any pending truncation is applied before doing anything
1874  * that may depend on it.
1875  */
1876 void __ceph_do_pending_vmtruncate(struct inode *inode)
1877 {
1878         struct ceph_inode_info *ci = ceph_inode(inode);
1879         u64 to;
1880         int wrbuffer_refs, finish = 0;
1881
1882         mutex_lock(&ci->i_truncate_mutex);
1883 retry:
1884         spin_lock(&ci->i_ceph_lock);
1885         if (ci->i_truncate_pending == 0) {
1886                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1887                 spin_unlock(&ci->i_ceph_lock);
1888                 mutex_unlock(&ci->i_truncate_mutex);
1889                 return;
1890         }
1891
1892         /*
1893          * make sure any dirty snapped pages are flushed before we
1894          * possibly truncate them.. so write AND block!
1895          */
1896         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1897                 spin_unlock(&ci->i_ceph_lock);
1898                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1899                      inode);
1900                 filemap_write_and_wait_range(&inode->i_data, 0,
1901                                              inode->i_sb->s_maxbytes);
1902                 goto retry;
1903         }
1904
1905         /* there should be no reader or writer */
1906         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1907
1908         to = ci->i_truncate_size;
1909         wrbuffer_refs = ci->i_wrbuffer_ref;
1910         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1911              ci->i_truncate_pending, to);
1912         spin_unlock(&ci->i_ceph_lock);
1913
1914         truncate_pagecache(inode, to);
1915
1916         spin_lock(&ci->i_ceph_lock);
1917         if (to == ci->i_truncate_size) {
1918                 ci->i_truncate_pending = 0;
1919                 finish = 1;
1920         }
1921         spin_unlock(&ci->i_ceph_lock);
1922         if (!finish)
1923                 goto retry;
1924
1925         mutex_unlock(&ci->i_truncate_mutex);
1926
1927         if (wrbuffer_refs == 0)
1928                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1929
1930         wake_up_all(&ci->i_cap_wq);
1931 }
1932
1933 /*
1934  * symlinks
1935  */
1936 static const struct inode_operations ceph_symlink_iops = {
1937         .get_link = simple_get_link,
1938         .setattr = ceph_setattr,
1939         .getattr = ceph_getattr,
1940         .listxattr = ceph_listxattr,
1941 };
1942
1943 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1944 {
1945         struct ceph_inode_info *ci = ceph_inode(inode);
1946         const unsigned int ia_valid = attr->ia_valid;
1947         struct ceph_mds_request *req;
1948         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1949         struct ceph_cap_flush *prealloc_cf;
1950         int issued;
1951         int release = 0, dirtied = 0;
1952         int mask = 0;
1953         int err = 0;
1954         int inode_dirty_flags = 0;
1955         bool lock_snap_rwsem = false;
1956
1957         prealloc_cf = ceph_alloc_cap_flush();
1958         if (!prealloc_cf)
1959                 return -ENOMEM;
1960
1961         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1962                                        USE_AUTH_MDS);
1963         if (IS_ERR(req)) {
1964                 ceph_free_cap_flush(prealloc_cf);
1965                 return PTR_ERR(req);
1966         }
1967
1968         spin_lock(&ci->i_ceph_lock);
1969         issued = __ceph_caps_issued(ci, NULL);
1970
1971         if (!ci->i_head_snapc &&
1972             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1973                 lock_snap_rwsem = true;
1974                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1975                         spin_unlock(&ci->i_ceph_lock);
1976                         down_read(&mdsc->snap_rwsem);
1977                         spin_lock(&ci->i_ceph_lock);
1978                         issued = __ceph_caps_issued(ci, NULL);
1979                 }
1980         }
1981
1982         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1983
1984         if (ia_valid & ATTR_UID) {
1985                 dout("setattr %p uid %d -> %d\n", inode,
1986                      from_kuid(&init_user_ns, inode->i_uid),
1987                      from_kuid(&init_user_ns, attr->ia_uid));
1988                 if (issued & CEPH_CAP_AUTH_EXCL) {
1989                         inode->i_uid = attr->ia_uid;
1990                         dirtied |= CEPH_CAP_AUTH_EXCL;
1991                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1992                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1993                         req->r_args.setattr.uid = cpu_to_le32(
1994                                 from_kuid(&init_user_ns, attr->ia_uid));
1995                         mask |= CEPH_SETATTR_UID;
1996                         release |= CEPH_CAP_AUTH_SHARED;
1997                 }
1998         }
1999         if (ia_valid & ATTR_GID) {
2000                 dout("setattr %p gid %d -> %d\n", inode,
2001                      from_kgid(&init_user_ns, inode->i_gid),
2002                      from_kgid(&init_user_ns, attr->ia_gid));
2003                 if (issued & CEPH_CAP_AUTH_EXCL) {
2004                         inode->i_gid = attr->ia_gid;
2005                         dirtied |= CEPH_CAP_AUTH_EXCL;
2006                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2007                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2008                         req->r_args.setattr.gid = cpu_to_le32(
2009                                 from_kgid(&init_user_ns, attr->ia_gid));
2010                         mask |= CEPH_SETATTR_GID;
2011                         release |= CEPH_CAP_AUTH_SHARED;
2012                 }
2013         }
2014         if (ia_valid & ATTR_MODE) {
2015                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2016                      attr->ia_mode);
2017                 if (issued & CEPH_CAP_AUTH_EXCL) {
2018                         inode->i_mode = attr->ia_mode;
2019                         dirtied |= CEPH_CAP_AUTH_EXCL;
2020                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2021                            attr->ia_mode != inode->i_mode) {
2022                         inode->i_mode = attr->ia_mode;
2023                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2024                         mask |= CEPH_SETATTR_MODE;
2025                         release |= CEPH_CAP_AUTH_SHARED;
2026                 }
2027         }
2028
2029         if (ia_valid & ATTR_ATIME) {
2030                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2031                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2032                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2033                 if (issued & CEPH_CAP_FILE_EXCL) {
2034                         ci->i_time_warp_seq++;
2035                         inode->i_atime = attr->ia_atime;
2036                         dirtied |= CEPH_CAP_FILE_EXCL;
2037                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2038                            timespec64_compare(&inode->i_atime,
2039                                             &attr->ia_atime) < 0) {
2040                         inode->i_atime = attr->ia_atime;
2041                         dirtied |= CEPH_CAP_FILE_WR;
2042                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2043                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2044                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2045                                                &attr->ia_atime);
2046                         mask |= CEPH_SETATTR_ATIME;
2047                         release |= CEPH_CAP_FILE_SHARED |
2048                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2049                 }
2050         }
2051         if (ia_valid & ATTR_MTIME) {
2052                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2053                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2054                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2055                 if (issued & CEPH_CAP_FILE_EXCL) {
2056                         ci->i_time_warp_seq++;
2057                         inode->i_mtime = attr->ia_mtime;
2058                         dirtied |= CEPH_CAP_FILE_EXCL;
2059                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2060                            timespec64_compare(&inode->i_mtime,
2061                                             &attr->ia_mtime) < 0) {
2062                         inode->i_mtime = attr->ia_mtime;
2063                         dirtied |= CEPH_CAP_FILE_WR;
2064                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2065                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2066                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2067                                                &attr->ia_mtime);
2068                         mask |= CEPH_SETATTR_MTIME;
2069                         release |= CEPH_CAP_FILE_SHARED |
2070                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2071                 }
2072         }
2073         if (ia_valid & ATTR_SIZE) {
2074                 dout("setattr %p size %lld -> %lld\n", inode,
2075                      inode->i_size, attr->ia_size);
2076                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2077                     attr->ia_size > inode->i_size) {
2078                         i_size_write(inode, attr->ia_size);
2079                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2080                         ci->i_reported_size = attr->ia_size;
2081                         dirtied |= CEPH_CAP_FILE_EXCL;
2082                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2083                            attr->ia_size != inode->i_size) {
2084                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2085                         req->r_args.setattr.old_size =
2086                                 cpu_to_le64(inode->i_size);
2087                         mask |= CEPH_SETATTR_SIZE;
2088                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2089                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2090                 }
2091         }
2092
2093         /* these do nothing */
2094         if (ia_valid & ATTR_CTIME) {
2095                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2096                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2097                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2098                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2099                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2100                      only ? "ctime only" : "ignored");
2101                 if (only) {
2102                         /*
2103                          * if kernel wants to dirty ctime but nothing else,
2104                          * we need to choose a cap to dirty under, or do
2105                          * a almost-no-op setattr
2106                          */
2107                         if (issued & CEPH_CAP_AUTH_EXCL)
2108                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2109                         else if (issued & CEPH_CAP_FILE_EXCL)
2110                                 dirtied |= CEPH_CAP_FILE_EXCL;
2111                         else if (issued & CEPH_CAP_XATTR_EXCL)
2112                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2113                         else
2114                                 mask |= CEPH_SETATTR_CTIME;
2115                 }
2116         }
2117         if (ia_valid & ATTR_FILE)
2118                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2119
2120         if (dirtied) {
2121                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2122                                                            &prealloc_cf);
2123                 inode->i_ctime = attr->ia_ctime;
2124         }
2125
2126         release &= issued;
2127         spin_unlock(&ci->i_ceph_lock);
2128         if (lock_snap_rwsem)
2129                 up_read(&mdsc->snap_rwsem);
2130
2131         if (inode_dirty_flags)
2132                 __mark_inode_dirty(inode, inode_dirty_flags);
2133
2134
2135         if (mask) {
2136                 req->r_inode = inode;
2137                 ihold(inode);
2138                 req->r_inode_drop = release;
2139                 req->r_args.setattr.mask = cpu_to_le32(mask);
2140                 req->r_num_caps = 1;
2141                 req->r_stamp = attr->ia_ctime;
2142                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2143         }
2144         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2145              ceph_cap_string(dirtied), mask);
2146
2147         ceph_mdsc_put_request(req);
2148         ceph_free_cap_flush(prealloc_cf);
2149
2150         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2151                 __ceph_do_pending_vmtruncate(inode);
2152
2153         return err;
2154 }
2155
2156 /*
2157  * setattr
2158  */
2159 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2160 {
2161         struct inode *inode = d_inode(dentry);
2162         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2163         int err;
2164
2165         if (ceph_snap(inode) != CEPH_NOSNAP)
2166                 return -EROFS;
2167
2168         err = setattr_prepare(dentry, attr);
2169         if (err != 0)
2170                 return err;
2171
2172         if ((attr->ia_valid & ATTR_SIZE) &&
2173             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2174                 return -EFBIG;
2175
2176         if ((attr->ia_valid & ATTR_SIZE) &&
2177             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2178                 return -EDQUOT;
2179
2180         err = __ceph_setattr(inode, attr);
2181
2182         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2183                 err = posix_acl_chmod(inode, attr->ia_mode);
2184
2185         return err;
2186 }
2187
2188 /*
2189  * Verify that we have a lease on the given mask.  If not,
2190  * do a getattr against an mds.
2191  */
2192 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2193                       int mask, bool force)
2194 {
2195         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2196         struct ceph_mds_client *mdsc = fsc->mdsc;
2197         struct ceph_mds_request *req;
2198         int mode;
2199         int err;
2200
2201         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2202                 dout("do_getattr inode %p SNAPDIR\n", inode);
2203                 return 0;
2204         }
2205
2206         dout("do_getattr inode %p mask %s mode 0%o\n",
2207              inode, ceph_cap_string(mask), inode->i_mode);
2208         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2209                 return 0;
2210
2211         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2212         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2213         if (IS_ERR(req))
2214                 return PTR_ERR(req);
2215         req->r_inode = inode;
2216         ihold(inode);
2217         req->r_num_caps = 1;
2218         req->r_args.getattr.mask = cpu_to_le32(mask);
2219         req->r_locked_page = locked_page;
2220         err = ceph_mdsc_do_request(mdsc, NULL, req);
2221         if (locked_page && err == 0) {
2222                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2223                 if (inline_version == 0) {
2224                         /* the reply is supposed to contain inline data */
2225                         err = -EINVAL;
2226                 } else if (inline_version == CEPH_INLINE_NONE) {
2227                         err = -ENODATA;
2228                 } else {
2229                         err = req->r_reply_info.targeti.inline_len;
2230                 }
2231         }
2232         ceph_mdsc_put_request(req);
2233         dout("do_getattr result=%d\n", err);
2234         return err;
2235 }
2236
2237
2238 /*
2239  * Check inode permissions.  We verify we have a valid value for
2240  * the AUTH cap, then call the generic handler.
2241  */
2242 int ceph_permission(struct inode *inode, int mask)
2243 {
2244         int err;
2245
2246         if (mask & MAY_NOT_BLOCK)
2247                 return -ECHILD;
2248
2249         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2250
2251         if (!err)
2252                 err = generic_permission(inode, mask);
2253         return err;
2254 }
2255
2256 /*
2257  * Get all attributes.  Hopefully somedata we'll have a statlite()
2258  * and can limit the fields we require to be accurate.
2259  */
2260 int ceph_getattr(const struct path *path, struct kstat *stat,
2261                  u32 request_mask, unsigned int flags)
2262 {
2263         struct inode *inode = d_inode(path->dentry);
2264         struct ceph_inode_info *ci = ceph_inode(inode);
2265         int err;
2266
2267         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2268         if (!err) {
2269                 generic_fillattr(inode, stat);
2270                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2271                 if (ceph_snap(inode) == CEPH_NOSNAP)
2272                         stat->dev = inode->i_sb->s_dev;
2273                 else
2274                         stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2275
2276                 if (S_ISDIR(inode->i_mode)) {
2277                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2278                                                 RBYTES))
2279                                 stat->size = ci->i_rbytes;
2280                         else
2281                                 stat->size = ci->i_files + ci->i_subdirs;
2282                         stat->blocks = 0;
2283                         stat->blksize = 65536;
2284                         /*
2285                          * Some applications rely on the number of st_nlink
2286                          * value on directories to be either 0 (if unlinked)
2287                          * or 2 + number of subdirectories.
2288                          */
2289                         if (stat->nlink == 1)
2290                                 /* '.' + '..' + subdirs */
2291                                 stat->nlink = 1 + 1 + ci->i_subdirs;
2292                 }
2293         }
2294         return err;
2295 }