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, 1);
 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_remove_caps(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                         ci->i_dir_pin = iinfo->dir_pin;
 881                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 882                 }
 883         }
 884
 885         /* xattrs */
 886         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 887         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 888             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 889                 if (ci->i_xattrs.blob)
 890                         ceph_buffer_put(ci->i_xattrs.blob);
 891                 ci->i_xattrs.blob = xattr_blob;
 892                 if (xattr_blob)
 893                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 894                                iinfo->xattr_data, iinfo->xattr_len);
 895                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 896                 ceph_forget_all_cached_acls(inode);
 897                 xattr_blob = NULL;
 898         }
 899
 900         /* finally update i_version */
 901         if (le64_to_cpu(info->version) > ci->i_version)
 902                 ci->i_version = le64_to_cpu(info->version);
 903
 904         inode->i_mapping->a_ops = &ceph_aops;
 905
 906         switch (inode->i_mode & S_IFMT) {
 907         case S_IFIFO:
 908         case S_IFBLK:
 909         case S_IFCHR:
 910         case S_IFSOCK:
 911                 inode->i_blkbits = PAGE_SHIFT;
 912                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 913                 inode->i_op = &ceph_file_iops;
 914                 break;
 915         case S_IFREG:
 916                 inode->i_op = &ceph_file_iops;
 917                 inode->i_fop = &ceph_file_fops;
 918                 break;
 919         case S_IFLNK:
 920                 inode->i_op = &ceph_symlink_iops;
 921                 if (!ci->i_symlink) {
 922                         u32 symlen = iinfo->symlink_len;
 923                         char *sym;
 924
 925                         spin_unlock(&ci->i_ceph_lock);
 926
 927                         if (symlen != i_size_read(inode)) {
 928                                 pr_err("fill_inode %llx.%llx BAD symlink "
 929                                         "size %lld\n", ceph_vinop(inode),
 930                                         i_size_read(inode));
 931                                 i_size_write(inode, symlen);
 932                                 inode->i_blocks = calc_inode_blocks(symlen);
 933                         }
 934
 935                         err = -ENOMEM;
 936                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 937                         if (!sym)
 938                                 goto out;
 939
 940                         spin_lock(&ci->i_ceph_lock);
 941                         if (!ci->i_symlink)
 942                                 ci->i_symlink = sym;
 943                         else
 944                                 kfree(sym); /* lost a race */
 945                 }
 946                 inode->i_link = ci->i_symlink;
 947                 break;
 948         case S_IFDIR:
 949                 inode->i_op = &ceph_dir_iops;
 950                 inode->i_fop = &ceph_dir_fops;
 951                 break;
 952         default:
 953                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 954                        ceph_vinop(inode), inode->i_mode);
 955         }
 956
 957         /* were we issued a capability? */
 958         if (info_caps) {
 959                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 960                         ceph_add_cap(inode, session,
 961                                      le64_to_cpu(info->cap.cap_id),
 962                                      cap_fmode, info_caps,
 963                                      le32_to_cpu(info->cap.wanted),
 964                                      le32_to_cpu(info->cap.seq),
 965                                      le32_to_cpu(info->cap.mseq),
 966                                      le64_to_cpu(info->cap.realm),
 967                                      info->cap.flags, &new_cap);
 968
 969                         /* set dir completion flag? */
 970                         if (S_ISDIR(inode->i_mode) &&
 971                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 972                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 973                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 974                             !__ceph_dir_is_complete(ci)) {
 975                                 dout(" marking %p complete (empty)\n", inode);
 976                                 i_size_write(inode, 0);
 977                                 __ceph_dir_set_complete(ci,
 978                                         atomic64_read(&ci->i_release_count),
 979                                         atomic64_read(&ci->i_ordered_count));
 980                         }
 981
 982                         wake = true;
 983                 } else {
 984                         dout(" %p got snap_caps %s\n", inode,
 985                              ceph_cap_string(info_caps));
 986                         ci->i_snap_caps |= info_caps;
 987                         if (cap_fmode >= 0)
 988                                 __ceph_get_fmode(ci, cap_fmode);
 989                 }
 990         } else if (cap_fmode >= 0) {
 991                 pr_warn("mds issued no caps on %llx.%llx\n",
 992                            ceph_vinop(inode));
 993                 __ceph_get_fmode(ci, cap_fmode);
 994         }
 995
 996         if (iinfo->inline_version > 0 &&
 997             iinfo->inline_version >= ci->i_inline_version) {
 998                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 999                 ci->i_inline_version = iinfo->inline_version;
1000                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1001                     (locked_page || (info_caps & cache_caps)))
1002                         fill_inline = true;
1003         }
1004
1005         spin_unlock(&ci->i_ceph_lock);
1006
1007         if (fill_inline)
1008                 ceph_fill_inline_data(inode, locked_page,
1009                                       iinfo->inline_data, iinfo->inline_len);
1010
1011         if (wake)
1012                 wake_up_all(&ci->i_cap_wq);
1013
1014         /* queue truncate if we saw i_size decrease */
1015         if (queue_trunc)
1016                 ceph_queue_vmtruncate(inode);
1017
1018         /* populate frag tree */
1019         if (S_ISDIR(inode->i_mode))
1020                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1021
1022         /* update delegation info? */
1023         if (dirinfo)
1024                 ceph_fill_dirfrag(inode, dirinfo);
1025
1026         err = 0;
1027 out:
1028         if (new_cap)
1029                 ceph_put_cap(mdsc, new_cap);
1030         if (xattr_blob)
1031                 ceph_buffer_put(xattr_blob);
1032         ceph_put_string(pool_ns);
1033         return err;
1034 }
1035
1036 /*
1037  * caller should hold session s_mutex.
1038  */
1039 static void update_dentry_lease(struct dentry *dentry,
1040                                 struct ceph_mds_reply_lease *lease,
1041                                 struct ceph_mds_session *session,
1042                                 unsigned long from_time,
1043                                 struct ceph_vino *tgt_vino,
1044                                 struct ceph_vino *dir_vino)
1045 {
1046         struct ceph_dentry_info *di = ceph_dentry(dentry);
1047         long unsigned duration = le32_to_cpu(lease->duration_ms);
1048         long unsigned ttl = from_time + (duration * HZ) / 1000;
1049         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1050         struct inode *dir;
1051         struct ceph_mds_session *old_lease_session = NULL;
1052
1053         /*
1054          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1055          * we expect a negative dentry.
1056          */
1057         if (!tgt_vino && d_really_is_positive(dentry))
1058                 return;
1059
1060         if (tgt_vino && (d_really_is_negative(dentry) ||
1061                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1062                 return;
1063
1064         spin_lock(&dentry->d_lock);
1065         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1066              dentry, duration, ttl);
1067
1068         dir = d_inode(dentry->d_parent);
1069
1070         /* make sure parent matches dir_vino */
1071         if (!ceph_ino_compare(dir, dir_vino))
1072                 goto out_unlock;
1073
1074         /* only track leases on regular dentries */
1075         if (ceph_snap(dir) != CEPH_NOSNAP)
1076                 goto out_unlock;
1077
1078         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1079         if (duration == 0) {
1080                 __ceph_dentry_dir_lease_touch(di);
1081                 goto out_unlock;
1082         }
1083
1084         if (di->lease_gen == session->s_cap_gen &&
1085             time_before(ttl, di->time))
1086                 goto out_unlock;  /* we already have a newer lease. */
1087
1088         if (di->lease_session && di->lease_session != session) {
1089                 old_lease_session = di->lease_session;
1090                 di->lease_session = NULL;
1091         }
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
1101         __ceph_dentry_lease_touch(di);
1102 out_unlock:
1103         spin_unlock(&dentry->d_lock);
1104         if (old_lease_session)
1105                 ceph_put_mds_session(old_lease_session);
1106 }
1107
1108 /*
1109  * splice a dentry to an inode.
1110  * caller must hold directory i_mutex for this to be safe.
1111  */
1112 static int splice_dentry(struct dentry **pdn, struct inode *in)
1113 {
1114         struct dentry *dn = *pdn;
1115         struct dentry *realdn;
1116
1117         BUG_ON(d_inode(dn));
1118
1119         if (S_ISDIR(in->i_mode)) {
1120                 /* If inode is directory, d_splice_alias() below will remove
1121                  * 'realdn' from its origin parent. We need to ensure that
1122                  * origin parent's readdir cache will not reference 'realdn'
1123                  */
1124                 realdn = d_find_any_alias(in);
1125                 if (realdn) {
1126                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1127                         spin_lock(&realdn->d_lock);
1128
1129                         realdn->d_op->d_prune(realdn);
1130
1131                         di->time = jiffies;
1132                         di->lease_shared_gen = 0;
1133                         di->offset = 0;
1134
1135                         spin_unlock(&realdn->d_lock);
1136                         dput(realdn);
1137                 }
1138         }
1139
1140         /* dn must be unhashed */
1141         if (!d_unhashed(dn))
1142                 d_drop(dn);
1143         realdn = d_splice_alias(in, dn);
1144         if (IS_ERR(realdn)) {
1145                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1146                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1147                 return PTR_ERR(realdn);
1148         }
1149
1150         if (realdn) {
1151                 dout("dn %p (%d) spliced with %p (%d) "
1152                      "inode %p ino %llx.%llx\n",
1153                      dn, d_count(dn),
1154                      realdn, d_count(realdn),
1155                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1156                 dput(dn);
1157                 *pdn = realdn;
1158         } else {
1159                 BUG_ON(!ceph_dentry(dn));
1160                 dout("dn %p attached to %p ino %llx.%llx\n",
1161                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1162         }
1163         return 0;
1164 }
1165
1166 /*
1167  * Incorporate results into the local cache.  This is either just
1168  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1169  * after a lookup).
1170  *
1171  * A reply may contain
1172  *         a directory inode along with a dentry.
1173  *  and/or a target inode
1174  *
1175  * Called with snap_rwsem (read).
1176  */
1177 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1178 {
1179         struct ceph_mds_session *session = req->r_session;
1180         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1181         struct inode *in = NULL;
1182         struct ceph_vino tvino, dvino;
1183         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1184         int err = 0;
1185
1186         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1187              rinfo->head->is_dentry, rinfo->head->is_target);
1188
1189         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1190                 dout("fill_trace reply is empty!\n");
1191                 if (rinfo->head->result == 0 && req->r_parent)
1192                         ceph_invalidate_dir_request(req);
1193                 return 0;
1194         }
1195
1196         if (rinfo->head->is_dentry) {
1197                 struct inode *dir = req->r_parent;
1198
1199                 if (dir) {
1200                         err = fill_inode(dir, NULL,
1201                                          &rinfo->diri, rinfo->dirfrag,
1202                                          session, req->r_request_started, -1,
1203                                          &req->r_caps_reservation);
1204                         if (err < 0)
1205                                 goto done;
1206                 } else {
1207                         WARN_ON_ONCE(1);
1208                 }
1209
1210                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1211                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1212                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1213                         struct qstr dname;
1214                         struct dentry *dn, *parent;
1215
1216                         BUG_ON(!rinfo->head->is_target);
1217                         BUG_ON(req->r_dentry);
1218
1219                         parent = d_find_any_alias(dir);
1220                         BUG_ON(!parent);
1221
1222                         dname.name = rinfo->dname;
1223                         dname.len = rinfo->dname_len;
1224                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1225                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1226                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1227 retry_lookup:
1228                         dn = d_lookup(parent, &dname);
1229                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1230                              parent, dname.len, dname.name, dn);
1231
1232                         if (!dn) {
1233                                 dn = d_alloc(parent, &dname);
1234                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1235                                      dname.len, dname.name, dn);
1236                                 if (!dn) {
1237                                         dput(parent);
1238                                         err = -ENOMEM;
1239                                         goto done;
1240                                 }
1241                                 err = 0;
1242                         } else if (d_really_is_positive(dn) &&
1243                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1244                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1245                                 dout(" dn %p points to wrong inode %p\n",
1246                                      dn, d_inode(dn));
1247                                 ceph_dir_clear_ordered(dir);
1248                                 d_delete(dn);
1249                                 dput(dn);
1250                                 goto retry_lookup;
1251                         }
1252
1253                         req->r_dentry = dn;
1254                         dput(parent);
1255                 }
1256         }
1257
1258         if (rinfo->head->is_target) {
1259                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1260                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1261
1262                 in = ceph_get_inode(sb, tvino);
1263                 if (IS_ERR(in)) {
1264                         err = PTR_ERR(in);
1265                         goto done;
1266                 }
1267                 req->r_target_inode = in;
1268
1269                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1270                                 session, req->r_request_started,
1271                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1272                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1273                                 &req->r_caps_reservation);
1274                 if (err < 0) {
1275                         pr_err("fill_inode badness %p %llx.%llx\n",
1276                                 in, ceph_vinop(in));
1277                         goto done;
1278                 }
1279         }
1280
1281         /*
1282          * ignore null lease/binding on snapdir ENOENT, or else we
1283          * will have trouble splicing in the virtual snapdir later
1284          */
1285         if (rinfo->head->is_dentry &&
1286             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1287             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1288             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1289                                                fsc->mount_options->snapdir_name,
1290                                                req->r_dentry->d_name.len))) {
1291                 /*
1292                  * lookup link rename   : null -> possibly existing inode
1293                  * mknod symlink mkdir  : null -> new inode
1294                  * unlink               : linked -> null
1295                  */
1296                 struct inode *dir = req->r_parent;
1297                 struct dentry *dn = req->r_dentry;
1298                 bool have_dir_cap, have_lease;
1299
1300                 BUG_ON(!dn);
1301                 BUG_ON(!dir);
1302                 BUG_ON(d_inode(dn->d_parent) != dir);
1303
1304                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1305                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1306
1307                 BUG_ON(ceph_ino(dir) != dvino.ino);
1308                 BUG_ON(ceph_snap(dir) != dvino.snap);
1309
1310                 /* do we have a lease on the whole dir? */
1311                 have_dir_cap =
1312                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1313                          CEPH_CAP_FILE_SHARED);
1314
1315                 /* do we have a dn lease? */
1316                 have_lease = have_dir_cap ||
1317                         le32_to_cpu(rinfo->dlease->duration_ms);
1318                 if (!have_lease)
1319                         dout("fill_trace  no dentry lease or dir cap\n");
1320
1321                 /* rename? */
1322                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1323                         struct inode *olddir = req->r_old_dentry_dir;
1324                         BUG_ON(!olddir);
1325
1326                         dout(" src %p '%pd' dst %p '%pd'\n",
1327                              req->r_old_dentry,
1328                              req->r_old_dentry,
1329                              dn, dn);
1330                         dout("fill_trace doing d_move %p -> %p\n",
1331                              req->r_old_dentry, dn);
1332
1333                         /* d_move screws up sibling dentries' offsets */
1334                         ceph_dir_clear_ordered(dir);
1335                         ceph_dir_clear_ordered(olddir);
1336
1337                         d_move(req->r_old_dentry, dn);
1338                         dout(" src %p '%pd' dst %p '%pd'\n",
1339                              req->r_old_dentry,
1340                              req->r_old_dentry,
1341                              dn, dn);
1342
1343                         /* ensure target dentry is invalidated, despite
1344                            rehashing bug in vfs_rename_dir */
1345                         ceph_invalidate_dentry_lease(dn);
1346
1347                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1348                              ceph_dentry(req->r_old_dentry)->offset);
1349
1350                         /* swap r_dentry and r_old_dentry in case that
1351                          * splice_dentry() gets called later. This is safe
1352                          * because no other place will use them */
1353                         req->r_dentry = req->r_old_dentry;
1354                         req->r_old_dentry = dn;
1355                         dn = req->r_dentry;
1356                 }
1357
1358                 /* null dentry? */
1359                 if (!rinfo->head->is_target) {
1360                         dout("fill_trace null dentry\n");
1361                         if (d_really_is_positive(dn)) {
1362                                 dout("d_delete %p\n", dn);
1363                                 ceph_dir_clear_ordered(dir);
1364                                 d_delete(dn);
1365                         } else if (have_lease) {
1366                                 if (d_unhashed(dn))
1367                                         d_add(dn, NULL);
1368                                 update_dentry_lease(dn, rinfo->dlease,
1369                                                     session,
1370                                                     req->r_request_started,
1371                                                     NULL, &dvino);
1372                         }
1373                         goto done;
1374                 }
1375
1376                 /* attach proper inode */
1377                 if (d_really_is_negative(dn)) {
1378                         ceph_dir_clear_ordered(dir);
1379                         ihold(in);
1380                         err = splice_dentry(&req->r_dentry, in);
1381                         if (err < 0)
1382                                 goto done;
1383                         dn = req->r_dentry;  /* may have spliced */
1384                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1385                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1386                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1387                              ceph_vinop(in));
1388                         d_invalidate(dn);
1389                         have_lease = false;
1390                 }
1391
1392                 if (have_lease) {
1393                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1394                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1395                         update_dentry_lease(dn, rinfo->dlease, session,
1396                                             req->r_request_started,
1397                                             &tvino, &dvino);
1398                 }
1399                 dout(" final dn %p\n", dn);
1400         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1401                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1402                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1403                 struct inode *dir = req->r_parent;
1404
1405                 /* fill out a snapdir LOOKUPSNAP dentry */
1406                 BUG_ON(!dir);
1407                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1408                 BUG_ON(!req->r_dentry);
1409                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1410                 ceph_dir_clear_ordered(dir);
1411                 ihold(in);
1412                 err = splice_dentry(&req->r_dentry, in);
1413                 if (err < 0)
1414                         goto done;
1415         } else if (rinfo->head->is_dentry) {
1416                 struct ceph_vino *ptvino = NULL;
1417
1418                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1419                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1420                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1421                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1422
1423                         if (rinfo->head->is_target) {
1424                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1425                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1426                                 ptvino = &tvino;
1427                         }
1428
1429                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1430                                 session, req->r_request_started, ptvino,
1431                                 &dvino);
1432                 } else {
1433                         dout("%s: no dentry lease or dir cap\n", __func__);
1434                 }
1435         }
1436 done:
1437         dout("fill_trace done err=%d\n", err);
1438         return err;
1439 }
1440
1441 /*
1442  * Prepopulate our cache with readdir results, leases, etc.
1443  */
1444 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1445                                            struct ceph_mds_session *session)
1446 {
1447         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1448         int i, err = 0;
1449
1450         for (i = 0; i < rinfo->dir_nr; i++) {
1451                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1452                 struct ceph_vino vino;
1453                 struct inode *in;
1454                 int rc;
1455
1456                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1457                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1458
1459                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1460                 if (IS_ERR(in)) {
1461                         err = PTR_ERR(in);
1462                         dout("new_inode badness got %d\n", err);
1463                         continue;
1464                 }
1465                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1466                                 req->r_request_started, -1,
1467                                 &req->r_caps_reservation);
1468                 if (rc < 0) {
1469                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1470                         err = rc;
1471                 }
1472                 iput(in);
1473         }
1474
1475         return err;
1476 }
1477
1478 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1479 {
1480         if (ctl->page) {
1481                 kunmap(ctl->page);
1482                 put_page(ctl->page);
1483                 ctl->page = NULL;
1484         }
1485 }
1486
1487 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1488                               struct ceph_readdir_cache_control *ctl,
1489                               struct ceph_mds_request *req)
1490 {
1491         struct ceph_inode_info *ci = ceph_inode(dir);
1492         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1493         unsigned idx = ctl->index % nsize;
1494         pgoff_t pgoff = ctl->index / nsize;
1495
1496         if (!ctl->page || pgoff != page_index(ctl->page)) {
1497                 ceph_readdir_cache_release(ctl);
1498                 if (idx == 0)
1499                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1500                 else
1501                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1502                 if (!ctl->page) {
1503                         ctl->index = -1;
1504                         return idx == 0 ? -ENOMEM : 0;
1505                 }
1506                 /* reading/filling the cache are serialized by
1507                  * i_mutex, no need to use page lock */
1508                 unlock_page(ctl->page);
1509                 ctl->dentries = kmap(ctl->page);
1510                 if (idx == 0)
1511                         memset(ctl->dentries, 0, PAGE_SIZE);
1512         }
1513
1514         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1515             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1516                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1517                 ctl->dentries[idx] = dn;
1518                 ctl->index++;
1519         } else {
1520                 dout("disable readdir cache\n");
1521                 ctl->index = -1;
1522         }
1523         return 0;
1524 }
1525
1526 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1527                              struct ceph_mds_session *session)
1528 {
1529         struct dentry *parent = req->r_dentry;
1530         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1531         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1532         struct qstr dname;
1533         struct dentry *dn;
1534         struct inode *in;
1535         int err = 0, skipped = 0, ret, i;
1536         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1537         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1538         u32 last_hash = 0;
1539         u32 fpos_offset;
1540         struct ceph_readdir_cache_control cache_ctl = {};
1541
1542         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1543                 return readdir_prepopulate_inodes_only(req, session);
1544
1545         if (rinfo->hash_order) {
1546                 if (req->r_path2) {
1547                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1548                                                   req->r_path2,
1549                                                   strlen(req->r_path2));
1550                         last_hash = ceph_frag_value(last_hash);
1551                 } else if (rinfo->offset_hash) {
1552                         /* mds understands offset_hash */
1553                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1554                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1555                 }
1556         }
1557
1558         if (rinfo->dir_dir &&
1559             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1560                 dout("readdir_prepopulate got new frag %x -> %x\n",
1561                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1562                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1563                 if (!rinfo->hash_order)
1564                         req->r_readdir_offset = 2;
1565         }
1566
1567         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1568                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1569                      rinfo->dir_nr, parent);
1570         } else {
1571                 dout("readdir_prepopulate %d items under dn %p\n",
1572                      rinfo->dir_nr, parent);
1573                 if (rinfo->dir_dir)
1574                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1575
1576                 if (ceph_frag_is_leftmost(frag) &&
1577                     req->r_readdir_offset == 2 &&
1578                     !(rinfo->hash_order && last_hash)) {
1579                         /* note dir version at start of readdir so we can
1580                          * tell if any dentries get dropped */
1581                         req->r_dir_release_cnt =
1582                                 atomic64_read(&ci->i_release_count);
1583                         req->r_dir_ordered_cnt =
1584                                 atomic64_read(&ci->i_ordered_count);
1585                         req->r_readdir_cache_idx = 0;
1586                 }
1587         }
1588
1589         cache_ctl.index = req->r_readdir_cache_idx;
1590         fpos_offset = req->r_readdir_offset;
1591
1592         /* FIXME: release caps/leases if error occurs */
1593         for (i = 0; i < rinfo->dir_nr; i++) {
1594                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1595                 struct ceph_vino tvino, dvino;
1596
1597                 dname.name = rde->name;
1598                 dname.len = rde->name_len;
1599                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1600
1601                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1602                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1603
1604                 if (rinfo->hash_order) {
1605                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1606                                                  rde->name, rde->name_len);
1607                         hash = ceph_frag_value(hash);
1608                         if (hash != last_hash)
1609                                 fpos_offset = 2;
1610                         last_hash = hash;
1611                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1612                 } else {
1613                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1614                 }
1615
1616 retry_lookup:
1617                 dn = d_lookup(parent, &dname);
1618                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1619                      parent, dname.len, dname.name, dn);
1620
1621                 if (!dn) {
1622                         dn = d_alloc(parent, &dname);
1623                         dout("d_alloc %p '%.*s' = %p\n", parent,
1624                              dname.len, dname.name, dn);
1625                         if (!dn) {
1626                                 dout("d_alloc badness\n");
1627                                 err = -ENOMEM;
1628                                 goto out;
1629                         }
1630                 } else if (d_really_is_positive(dn) &&
1631                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1632                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1633                         struct ceph_dentry_info *di = ceph_dentry(dn);
1634                         dout(" dn %p points to wrong inode %p\n",
1635                              dn, d_inode(dn));
1636
1637                         spin_lock(&dn->d_lock);
1638                         if (di->offset > 0 &&
1639                             di->lease_shared_gen ==
1640                             atomic_read(&ci->i_shared_gen)) {
1641                                 __ceph_dir_clear_ordered(ci);
1642                                 di->offset = 0;
1643                         }
1644                         spin_unlock(&dn->d_lock);
1645
1646                         d_delete(dn);
1647                         dput(dn);
1648                         goto retry_lookup;
1649                 }
1650
1651                 /* inode */
1652                 if (d_really_is_positive(dn)) {
1653                         in = d_inode(dn);
1654                 } else {
1655                         in = ceph_get_inode(parent->d_sb, tvino);
1656                         if (IS_ERR(in)) {
1657                                 dout("new_inode badness\n");
1658                                 d_drop(dn);
1659                                 dput(dn);
1660                                 err = PTR_ERR(in);
1661                                 goto out;
1662                         }
1663                 }
1664
1665                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1666                                  req->r_request_started, -1,
1667                                  &req->r_caps_reservation);
1668                 if (ret < 0) {
1669                         pr_err("fill_inode badness on %p\n", in);
1670                         if (d_really_is_negative(dn))
1671                                 iput(in);
1672                         d_drop(dn);
1673                         err = ret;
1674                         goto next_item;
1675                 }
1676
1677                 if (d_really_is_negative(dn)) {
1678                         if (ceph_security_xattr_deadlock(in)) {
1679                                 dout(" skip splicing dn %p to inode %p"
1680                                      " (security xattr deadlock)\n", dn, in);
1681                                 iput(in);
1682                                 skipped++;
1683                                 goto next_item;
1684                         }
1685
1686                         err = splice_dentry(&dn, in);
1687                         if (err < 0)
1688                                 goto next_item;
1689                 }
1690
1691                 ceph_dentry(dn)->offset = rde->offset;
1692
1693                 dvino = ceph_vino(d_inode(parent));
1694                 update_dentry_lease(dn, rde->lease, req->r_session,
1695                                     req->r_request_started, &tvino, &dvino);
1696
1697                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1698                         ret = fill_readdir_cache(d_inode(parent), dn,
1699                                                  &cache_ctl, req);
1700                         if (ret < 0)
1701                                 err = ret;
1702                 }
1703 next_item:
1704                 dput(dn);
1705         }
1706 out:
1707         if (err == 0 && skipped == 0) {
1708                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1709                 req->r_readdir_cache_idx = cache_ctl.index;
1710         }
1711         ceph_readdir_cache_release(&cache_ctl);
1712         dout("readdir_prepopulate done\n");
1713         return err;
1714 }
1715
1716 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1717 {
1718         struct ceph_inode_info *ci = ceph_inode(inode);
1719         bool ret;
1720
1721         spin_lock(&ci->i_ceph_lock);
1722         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1723         i_size_write(inode, size);
1724         inode->i_blocks = calc_inode_blocks(size);
1725
1726         ret = __ceph_should_report_size(ci);
1727
1728         spin_unlock(&ci->i_ceph_lock);
1729         return ret;
1730 }
1731
1732 /*
1733  * Write back inode data in a worker thread.  (This can't be done
1734  * in the message handler context.)
1735  */
1736 void ceph_queue_writeback(struct inode *inode)
1737 {
1738         ihold(inode);
1739         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1740                        &ceph_inode(inode)->i_wb_work)) {
1741                 dout("ceph_queue_writeback %p\n", inode);
1742         } else {
1743                 dout("ceph_queue_writeback %p failed\n", inode);
1744                 iput(inode);
1745         }
1746 }
1747
1748 static void ceph_writeback_work(struct work_struct *work)
1749 {
1750         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1751                                                   i_wb_work);
1752         struct inode *inode = &ci->vfs_inode;
1753
1754         dout("writeback %p\n", inode);
1755         filemap_fdatawrite(&inode->i_data);
1756         iput(inode);
1757 }
1758
1759 /*
1760  * queue an async invalidation
1761  */
1762 void ceph_queue_invalidate(struct inode *inode)
1763 {
1764         ihold(inode);
1765         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1766                        &ceph_inode(inode)->i_pg_inv_work)) {
1767                 dout("ceph_queue_invalidate %p\n", inode);
1768         } else {
1769                 dout("ceph_queue_invalidate %p failed\n", inode);
1770                 iput(inode);
1771         }
1772 }
1773
1774 /*
1775  * Invalidate inode pages in a worker thread.  (This can't be done
1776  * in the message handler context.)
1777  */
1778 static void ceph_invalidate_work(struct work_struct *work)
1779 {
1780         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1781                                                   i_pg_inv_work);
1782         struct inode *inode = &ci->vfs_inode;
1783         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1784         u32 orig_gen;
1785         int check = 0;
1786
1787         mutex_lock(&ci->i_truncate_mutex);
1788
1789         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1790                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1791                                     inode, ceph_ino(inode));
1792                 mapping_set_error(inode->i_mapping, -EIO);
1793                 truncate_pagecache(inode, 0);
1794                 mutex_unlock(&ci->i_truncate_mutex);
1795                 goto out;
1796         }
1797
1798         spin_lock(&ci->i_ceph_lock);
1799         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1800              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1801         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1802                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1803                         check = 1;
1804                 spin_unlock(&ci->i_ceph_lock);
1805                 mutex_unlock(&ci->i_truncate_mutex);
1806                 goto out;
1807         }
1808         orig_gen = ci->i_rdcache_gen;
1809         spin_unlock(&ci->i_ceph_lock);
1810
1811         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1812                 pr_err("invalidate_pages %p fails\n", inode);
1813         }
1814
1815         spin_lock(&ci->i_ceph_lock);
1816         if (orig_gen == ci->i_rdcache_gen &&
1817             orig_gen == ci->i_rdcache_revoking) {
1818                 dout("invalidate_pages %p gen %d successful\n", inode,
1819                      ci->i_rdcache_gen);
1820                 ci->i_rdcache_revoking--;
1821                 check = 1;
1822         } else {
1823                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1824                      inode, orig_gen, ci->i_rdcache_gen,
1825                      ci->i_rdcache_revoking);
1826                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1827                         check = 1;
1828         }
1829         spin_unlock(&ci->i_ceph_lock);
1830         mutex_unlock(&ci->i_truncate_mutex);
1831 out:
1832         if (check)
1833                 ceph_check_caps(ci, 0, NULL);
1834         iput(inode);
1835 }
1836
1837
1838 /*
1839  * called by trunc_wq;
1840  *
1841  * We also truncate in a separate thread as well.
1842  */
1843 static void ceph_vmtruncate_work(struct work_struct *work)
1844 {
1845         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1846                                                   i_vmtruncate_work);
1847         struct inode *inode = &ci->vfs_inode;
1848
1849         dout("vmtruncate_work %p\n", inode);
1850         __ceph_do_pending_vmtruncate(inode);
1851         iput(inode);
1852 }
1853
1854 /*
1855  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1856  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1857  */
1858 void ceph_queue_vmtruncate(struct inode *inode)
1859 {
1860         struct ceph_inode_info *ci = ceph_inode(inode);
1861
1862         ihold(inode);
1863
1864         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1865                        &ci->i_vmtruncate_work)) {
1866                 dout("ceph_queue_vmtruncate %p\n", inode);
1867         } else {
1868                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1869                      inode, ci->i_truncate_pending);
1870                 iput(inode);
1871         }
1872 }
1873
1874 /*
1875  * Make sure any pending truncation is applied before doing anything
1876  * that may depend on it.
1877  */
1878 void __ceph_do_pending_vmtruncate(struct inode *inode)
1879 {
1880         struct ceph_inode_info *ci = ceph_inode(inode);
1881         u64 to;
1882         int wrbuffer_refs, finish = 0;
1883
1884         mutex_lock(&ci->i_truncate_mutex);
1885 retry:
1886         spin_lock(&ci->i_ceph_lock);
1887         if (ci->i_truncate_pending == 0) {
1888                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1889                 spin_unlock(&ci->i_ceph_lock);
1890                 mutex_unlock(&ci->i_truncate_mutex);
1891                 return;
1892         }
1893
1894         /*
1895          * make sure any dirty snapped pages are flushed before we
1896          * possibly truncate them.. so write AND block!
1897          */
1898         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1899                 spin_unlock(&ci->i_ceph_lock);
1900                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1901                      inode);
1902                 filemap_write_and_wait_range(&inode->i_data, 0,
1903                                              inode->i_sb->s_maxbytes);
1904                 goto retry;
1905         }
1906
1907         /* there should be no reader or writer */
1908         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1909
1910         to = ci->i_truncate_size;
1911         wrbuffer_refs = ci->i_wrbuffer_ref;
1912         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1913              ci->i_truncate_pending, to);
1914         spin_unlock(&ci->i_ceph_lock);
1915
1916         truncate_pagecache(inode, to);
1917
1918         spin_lock(&ci->i_ceph_lock);
1919         if (to == ci->i_truncate_size) {
1920                 ci->i_truncate_pending = 0;
1921                 finish = 1;
1922         }
1923         spin_unlock(&ci->i_ceph_lock);
1924         if (!finish)
1925                 goto retry;
1926
1927         mutex_unlock(&ci->i_truncate_mutex);
1928
1929         if (wrbuffer_refs == 0)
1930                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1931
1932         wake_up_all(&ci->i_cap_wq);
1933 }
1934
1935 /*
1936  * symlinks
1937  */
1938 static const struct inode_operations ceph_symlink_iops = {
1939         .get_link = simple_get_link,
1940         .setattr = ceph_setattr,
1941         .getattr = ceph_getattr,
1942         .listxattr = ceph_listxattr,
1943 };
1944
1945 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1946 {
1947         struct ceph_inode_info *ci = ceph_inode(inode);
1948         const unsigned int ia_valid = attr->ia_valid;
1949         struct ceph_mds_request *req;
1950         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1951         struct ceph_cap_flush *prealloc_cf;
1952         int issued;
1953         int release = 0, dirtied = 0;
1954         int mask = 0;
1955         int err = 0;
1956         int inode_dirty_flags = 0;
1957         bool lock_snap_rwsem = false;
1958
1959         prealloc_cf = ceph_alloc_cap_flush();
1960         if (!prealloc_cf)
1961                 return -ENOMEM;
1962
1963         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1964                                        USE_AUTH_MDS);
1965         if (IS_ERR(req)) {
1966                 ceph_free_cap_flush(prealloc_cf);
1967                 return PTR_ERR(req);
1968         }
1969
1970         spin_lock(&ci->i_ceph_lock);
1971         issued = __ceph_caps_issued(ci, NULL);
1972
1973         if (!ci->i_head_snapc &&
1974             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1975                 lock_snap_rwsem = true;
1976                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1977                         spin_unlock(&ci->i_ceph_lock);
1978                         down_read(&mdsc->snap_rwsem);
1979                         spin_lock(&ci->i_ceph_lock);
1980                         issued = __ceph_caps_issued(ci, NULL);
1981                 }
1982         }
1983
1984         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1985
1986         if (ia_valid & ATTR_UID) {
1987                 dout("setattr %p uid %d -> %d\n", inode,
1988                      from_kuid(&init_user_ns, inode->i_uid),
1989                      from_kuid(&init_user_ns, attr->ia_uid));
1990                 if (issued & CEPH_CAP_AUTH_EXCL) {
1991                         inode->i_uid = attr->ia_uid;
1992                         dirtied |= CEPH_CAP_AUTH_EXCL;
1993                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1994                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1995                         req->r_args.setattr.uid = cpu_to_le32(
1996                                 from_kuid(&init_user_ns, attr->ia_uid));
1997                         mask |= CEPH_SETATTR_UID;
1998                         release |= CEPH_CAP_AUTH_SHARED;
1999                 }
2000         }
2001         if (ia_valid & ATTR_GID) {
2002                 dout("setattr %p gid %d -> %d\n", inode,
2003                      from_kgid(&init_user_ns, inode->i_gid),
2004                      from_kgid(&init_user_ns, attr->ia_gid));
2005                 if (issued & CEPH_CAP_AUTH_EXCL) {
2006                         inode->i_gid = attr->ia_gid;
2007                         dirtied |= CEPH_CAP_AUTH_EXCL;
2008                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2009                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2010                         req->r_args.setattr.gid = cpu_to_le32(
2011                                 from_kgid(&init_user_ns, attr->ia_gid));
2012                         mask |= CEPH_SETATTR_GID;
2013                         release |= CEPH_CAP_AUTH_SHARED;
2014                 }
2015         }
2016         if (ia_valid & ATTR_MODE) {
2017                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2018                      attr->ia_mode);
2019                 if (issued & CEPH_CAP_AUTH_EXCL) {
2020                         inode->i_mode = attr->ia_mode;
2021                         dirtied |= CEPH_CAP_AUTH_EXCL;
2022                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2023                            attr->ia_mode != inode->i_mode) {
2024                         inode->i_mode = attr->ia_mode;
2025                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2026                         mask |= CEPH_SETATTR_MODE;
2027                         release |= CEPH_CAP_AUTH_SHARED;
2028                 }
2029         }
2030
2031         if (ia_valid & ATTR_ATIME) {
2032                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2033                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2034                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2035                 if (issued & CEPH_CAP_FILE_EXCL) {
2036                         ci->i_time_warp_seq++;
2037                         inode->i_atime = attr->ia_atime;
2038                         dirtied |= CEPH_CAP_FILE_EXCL;
2039                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2040                            timespec64_compare(&inode->i_atime,
2041                                             &attr->ia_atime) < 0) {
2042                         inode->i_atime = attr->ia_atime;
2043                         dirtied |= CEPH_CAP_FILE_WR;
2044                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2045                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2046                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2047                                                &attr->ia_atime);
2048                         mask |= CEPH_SETATTR_ATIME;
2049                         release |= CEPH_CAP_FILE_SHARED |
2050                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2051                 }
2052         }
2053         if (ia_valid & ATTR_MTIME) {
2054                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2055                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2056                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2057                 if (issued & CEPH_CAP_FILE_EXCL) {
2058                         ci->i_time_warp_seq++;
2059                         inode->i_mtime = attr->ia_mtime;
2060                         dirtied |= CEPH_CAP_FILE_EXCL;
2061                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2062                            timespec64_compare(&inode->i_mtime,
2063                                             &attr->ia_mtime) < 0) {
2064                         inode->i_mtime = attr->ia_mtime;
2065                         dirtied |= CEPH_CAP_FILE_WR;
2066                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2067                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2068                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2069                                                &attr->ia_mtime);
2070                         mask |= CEPH_SETATTR_MTIME;
2071                         release |= CEPH_CAP_FILE_SHARED |
2072                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2073                 }
2074         }
2075         if (ia_valid & ATTR_SIZE) {
2076                 dout("setattr %p size %lld -> %lld\n", inode,
2077                      inode->i_size, attr->ia_size);
2078                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2079                     attr->ia_size > inode->i_size) {
2080                         i_size_write(inode, attr->ia_size);
2081                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2082                         ci->i_reported_size = attr->ia_size;
2083                         dirtied |= CEPH_CAP_FILE_EXCL;
2084                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2085                            attr->ia_size != inode->i_size) {
2086                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2087                         req->r_args.setattr.old_size =
2088                                 cpu_to_le64(inode->i_size);
2089                         mask |= CEPH_SETATTR_SIZE;
2090                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2091                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2092                 }
2093         }
2094
2095         /* these do nothing */
2096         if (ia_valid & ATTR_CTIME) {
2097                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2098                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2099                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2100                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2101                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2102                      only ? "ctime only" : "ignored");
2103                 if (only) {
2104                         /*
2105                          * if kernel wants to dirty ctime but nothing else,
2106                          * we need to choose a cap to dirty under, or do
2107                          * a almost-no-op setattr
2108                          */
2109                         if (issued & CEPH_CAP_AUTH_EXCL)
2110                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2111                         else if (issued & CEPH_CAP_FILE_EXCL)
2112                                 dirtied |= CEPH_CAP_FILE_EXCL;
2113                         else if (issued & CEPH_CAP_XATTR_EXCL)
2114                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2115                         else
2116                                 mask |= CEPH_SETATTR_CTIME;
2117                 }
2118         }
2119         if (ia_valid & ATTR_FILE)
2120                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2121
2122         if (dirtied) {
2123                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2124                                                            &prealloc_cf);
2125                 inode->i_ctime = attr->ia_ctime;
2126         }
2127
2128         release &= issued;
2129         spin_unlock(&ci->i_ceph_lock);
2130         if (lock_snap_rwsem)
2131                 up_read(&mdsc->snap_rwsem);
2132
2133         if (inode_dirty_flags)
2134                 __mark_inode_dirty(inode, inode_dirty_flags);
2135
2136
2137         if (mask) {
2138                 req->r_inode = inode;
2139                 ihold(inode);
2140                 req->r_inode_drop = release;
2141                 req->r_args.setattr.mask = cpu_to_le32(mask);
2142                 req->r_num_caps = 1;
2143                 req->r_stamp = attr->ia_ctime;
2144                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2145         }
2146         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2147              ceph_cap_string(dirtied), mask);
2148
2149         ceph_mdsc_put_request(req);
2150         ceph_free_cap_flush(prealloc_cf);
2151
2152         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2153                 __ceph_do_pending_vmtruncate(inode);
2154
2155         return err;
2156 }
2157
2158 /*
2159  * setattr
2160  */
2161 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2162 {
2163         struct inode *inode = d_inode(dentry);
2164         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2165         int err;
2166
2167         if (ceph_snap(inode) != CEPH_NOSNAP)
2168                 return -EROFS;
2169
2170         err = setattr_prepare(dentry, attr);
2171         if (err != 0)
2172                 return err;
2173
2174         if ((attr->ia_valid & ATTR_SIZE) &&
2175             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2176                 return -EFBIG;
2177
2178         if ((attr->ia_valid & ATTR_SIZE) &&
2179             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2180                 return -EDQUOT;
2181
2182         err = __ceph_setattr(inode, attr);
2183
2184         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2185                 err = posix_acl_chmod(inode, attr->ia_mode);
2186
2187         return err;
2188 }
2189
2190 /*
2191  * Verify that we have a lease on the given mask.  If not,
2192  * do a getattr against an mds.
2193  */
2194 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2195                       int mask, bool force)
2196 {
2197         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2198         struct ceph_mds_client *mdsc = fsc->mdsc;
2199         struct ceph_mds_request *req;
2200         int mode;
2201         int err;
2202
2203         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2204                 dout("do_getattr inode %p SNAPDIR\n", inode);
2205                 return 0;
2206         }
2207
2208         dout("do_getattr inode %p mask %s mode 0%o\n",
2209              inode, ceph_cap_string(mask), inode->i_mode);
2210         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2211                 return 0;
2212
2213         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2214         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2215         if (IS_ERR(req))
2216                 return PTR_ERR(req);
2217         req->r_inode = inode;
2218         ihold(inode);
2219         req->r_num_caps = 1;
2220         req->r_args.getattr.mask = cpu_to_le32(mask);
2221         req->r_locked_page = locked_page;
2222         err = ceph_mdsc_do_request(mdsc, NULL, req);
2223         if (locked_page && err == 0) {
2224                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2225                 if (inline_version == 0) {
2226                         /* the reply is supposed to contain inline data */
2227                         err = -EINVAL;
2228                 } else if (inline_version == CEPH_INLINE_NONE) {
2229                         err = -ENODATA;
2230                 } else {
2231                         err = req->r_reply_info.targeti.inline_len;
2232                 }
2233         }
2234         ceph_mdsc_put_request(req);
2235         dout("do_getattr result=%d\n", err);
2236         return err;
2237 }
2238
2239
2240 /*
2241  * Check inode permissions.  We verify we have a valid value for
2242  * the AUTH cap, then call the generic handler.
2243  */
2244 int ceph_permission(struct inode *inode, int mask)
2245 {
2246         int err;
2247
2248         if (mask & MAY_NOT_BLOCK)
2249                 return -ECHILD;
2250
2251         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2252
2253         if (!err)
2254                 err = generic_permission(inode, mask);
2255         return err;
2256 }
2257
2258 /*
2259  * Get all attributes.  Hopefully somedata we'll have a statlite()
2260  * and can limit the fields we require to be accurate.
2261  */
2262 int ceph_getattr(const struct path *path, struct kstat *stat,
2263                  u32 request_mask, unsigned int flags)
2264 {
2265         struct inode *inode = d_inode(path->dentry);
2266         struct ceph_inode_info *ci = ceph_inode(inode);
2267         int err;
2268
2269         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2270         if (!err) {
2271                 generic_fillattr(inode, stat);
2272                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2273                 if (ceph_snap(inode) == CEPH_NOSNAP)
2274                         stat->dev = inode->i_sb->s_dev;
2275                 else
2276                         stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2277
2278                 if (S_ISDIR(inode->i_mode)) {
2279                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2280                                                 RBYTES))
2281                                 stat->size = ci->i_rbytes;
2282                         else
2283                                 stat->size = ci->i_files + ci->i_subdirs;
2284                         stat->blocks = 0;
2285                         stat->blksize = 65536;
2286                         /*
2287                          * Some applications rely on the number of st_nlink
2288                          * value on directories to be either 0 (if unlinked)
2289                          * or 2 + number of subdirectories.
2290                          */
2291                         if (stat->nlink == 1)
2292                                 /* '.' + '..' + subdirs */
2293                                 stat->nlink = 1 + 1 + ci->i_subdirs;
2294                 }
2295         }
2296         return err;
2297 }