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 void ceph_free_inode(struct inode *inode)
 523 {
 524         struct ceph_inode_info *ci = ceph_inode(inode);
 525
 526         kfree(ci->i_symlink);
 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         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 570                 frag = rb_entry(n, struct ceph_inode_frag, node);
 571                 rb_erase(n, &ci->i_fragtree);
 572                 kfree(frag);
 573         }
 574         ci->i_fragtree_nsplits = 0;
 575
 576         __ceph_destroy_xattrs(ci);
 577         if (ci->i_xattrs.blob)
 578                 ceph_buffer_put(ci->i_xattrs.blob);
 579         if (ci->i_xattrs.prealloc_blob)
 580                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 581
 582         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 583 }
 584
 585 int ceph_drop_inode(struct inode *inode)
 586 {
 587         /*
 588          * Positve dentry and corresponding inode are always accompanied
 589          * in MDS reply. So no need to keep inode in the cache after
 590          * dropping all its aliases.
 591          */
 592         return 1;
 593 }
 594
 595 static inline blkcnt_t calc_inode_blocks(u64 size)
 596 {
 597         return (size + (1<<9) - 1) >> 9;
 598 }
 599
 600 /*
 601  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 602  * careful because either the client or MDS may have more up to date
 603  * info, depending on which capabilities are held, and whether
 604  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 605  * and size are monotonically increasing, except when utimes() or
 606  * truncate() increments the corresponding _seq values.)
 607  */
 608 int ceph_fill_file_size(struct inode *inode, int issued,
 609                         u32 truncate_seq, u64 truncate_size, u64 size)
 610 {
 611         struct ceph_inode_info *ci = ceph_inode(inode);
 612         int queue_trunc = 0;
 613
 614         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 615             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 616                 dout("size %lld -> %llu\n", inode->i_size, size);
 617                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 618                         pr_err("fill_file_size non-zero size for directory\n");
 619                         size = 0;
 620                 }
 621                 i_size_write(inode, size);
 622                 inode->i_blocks = calc_inode_blocks(size);
 623                 ci->i_reported_size = size;
 624                 if (truncate_seq != ci->i_truncate_seq) {
 625                         dout("truncate_seq %u -> %u\n",
 626                              ci->i_truncate_seq, truncate_seq);
 627                         ci->i_truncate_seq = truncate_seq;
 628
 629                         /* the MDS should have revoked these caps */
 630                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 631                                                CEPH_CAP_FILE_RD |
 632                                                CEPH_CAP_FILE_WR |
 633                                                CEPH_CAP_FILE_LAZYIO));
 634                         /*
 635                          * If we hold relevant caps, or in the case where we're
 636                          * not the only client referencing this file and we
 637                          * don't hold those caps, then we need to check whether
 638                          * the file is either opened or mmaped
 639                          */
 640                         if ((issued & (CEPH_CAP_FILE_CACHE|
 641                                        CEPH_CAP_FILE_BUFFER)) ||
 642                             mapping_mapped(inode->i_mapping) ||
 643                             __ceph_caps_file_wanted(ci)) {
 644                                 ci->i_truncate_pending++;
 645                                 queue_trunc = 1;
 646                         }
 647                 }
 648         }
 649         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 650             ci->i_truncate_size != truncate_size) {
 651                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 652                      truncate_size);
 653                 ci->i_truncate_size = truncate_size;
 654         }
 655
 656         if (queue_trunc)
 657                 ceph_fscache_invalidate(inode);
 658
 659         return queue_trunc;
 660 }
 661
 662 void ceph_fill_file_time(struct inode *inode, int issued,
 663                          u64 time_warp_seq, struct timespec64 *ctime,
 664                          struct timespec64 *mtime, struct timespec64 *atime)
 665 {
 666         struct ceph_inode_info *ci = ceph_inode(inode);
 667         int warn = 0;
 668
 669         if (issued & (CEPH_CAP_FILE_EXCL|
 670                       CEPH_CAP_FILE_WR|
 671                       CEPH_CAP_FILE_BUFFER|
 672                       CEPH_CAP_AUTH_EXCL|
 673                       CEPH_CAP_XATTR_EXCL)) {
 674                 if (ci->i_version == 0 ||
 675                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 676                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 677                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 678                              ctime->tv_sec, ctime->tv_nsec);
 679                         inode->i_ctime = *ctime;
 680                 }
 681                 if (ci->i_version == 0 ||
 682                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 683                         /* the MDS did a utimes() */
 684                         dout("mtime %lld.%09ld -> %lld.%09ld "
 685                              "tw %d -> %d\n",
 686                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 687                              mtime->tv_sec, mtime->tv_nsec,
 688                              ci->i_time_warp_seq, (int)time_warp_seq);
 689
 690                         inode->i_mtime = *mtime;
 691                         inode->i_atime = *atime;
 692                         ci->i_time_warp_seq = time_warp_seq;
 693                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 694                         /* nobody did utimes(); take the max */
 695                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 696                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 697                                      inode->i_mtime.tv_sec,
 698                                      inode->i_mtime.tv_nsec,
 699                                      mtime->tv_sec, mtime->tv_nsec);
 700                                 inode->i_mtime = *mtime;
 701                         }
 702                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 703                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 704                                      inode->i_atime.tv_sec,
 705                                      inode->i_atime.tv_nsec,
 706                                      atime->tv_sec, atime->tv_nsec);
 707                                 inode->i_atime = *atime;
 708                         }
 709                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 710                         /* we did a utimes(); ignore mds values */
 711                 } else {
 712                         warn = 1;
 713                 }
 714         } else {
 715                 /* we have no write|excl caps; whatever the MDS says is true */
 716                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 717                         inode->i_ctime = *ctime;
 718                         inode->i_mtime = *mtime;
 719                         inode->i_atime = *atime;
 720                         ci->i_time_warp_seq = time_warp_seq;
 721                 } else {
 722                         warn = 1;
 723                 }
 724         }
 725         if (warn) /* time_warp_seq shouldn't go backwards */
 726                 dout("%p mds time_warp_seq %llu < %u\n",
 727                      inode, time_warp_seq, ci->i_time_warp_seq);
 728 }
 729
 730 /*
 731  * Populate an inode based on info from mds.  May be called on new or
 732  * existing inodes.
 733  */
 734 static int fill_inode(struct inode *inode, struct page *locked_page,
 735                       struct ceph_mds_reply_info_in *iinfo,
 736                       struct ceph_mds_reply_dirfrag *dirinfo,
 737                       struct ceph_mds_session *session,
 738                       unsigned long ttl_from, int cap_fmode,
 739                       struct ceph_cap_reservation *caps_reservation)
 740 {
 741         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 742         struct ceph_mds_reply_inode *info = iinfo->in;
 743         struct ceph_inode_info *ci = ceph_inode(inode);
 744         int issued, new_issued, info_caps;
 745         struct timespec64 mtime, atime, ctime;
 746         struct ceph_buffer *xattr_blob = NULL;
 747         struct ceph_string *pool_ns = NULL;
 748         struct ceph_cap *new_cap = NULL;
 749         int err = 0;
 750         bool wake = false;
 751         bool queue_trunc = false;
 752         bool new_version = false;
 753         bool fill_inline = false;
 754
 755         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 756              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 757              ci->i_version);
 758
 759         info_caps = le32_to_cpu(info->cap.caps);
 760
 761         /* prealloc new cap struct */
 762         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
 763                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 764
 765         /*
 766          * prealloc xattr data, if it looks like we'll need it.  only
 767          * if len > 4 (meaning there are actually xattrs; the first 4
 768          * bytes are the xattr count).
 769          */
 770         if (iinfo->xattr_len > 4) {
 771                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 772                 if (!xattr_blob)
 773                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 774                                iinfo->xattr_len);
 775         }
 776
 777         if (iinfo->pool_ns_len > 0)
 778                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 779                                                      iinfo->pool_ns_len);
 780
 781         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 782                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 783
 784         spin_lock(&ci->i_ceph_lock);
 785
 786         /*
 787          * provided version will be odd if inode value is projected,
 788          * even if stable.  skip the update if we have newer stable
 789          * info (ours>=theirs, e.g. due to racing mds replies), unless
 790          * we are getting projected (unstable) info (in which case the
 791          * version is odd, and we want ours>theirs).
 792          *   us   them
 793          *   2    2     skip
 794          *   3    2     skip
 795          *   3    3     update
 796          */
 797         if (ci->i_version == 0 ||
 798             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 799              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 800                 new_version = true;
 801
 802         __ceph_caps_issued(ci, &issued);
 803         issued |= __ceph_caps_dirty(ci);
 804         new_issued = ~issued & info_caps;
 805
 806         /* update inode */
 807         inode->i_rdev = le32_to_cpu(info->rdev);
 808         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 809
 810         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 811
 812         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 813             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 814                 inode->i_mode = le32_to_cpu(info->mode);
 815                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 816                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 817                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 818                      from_kuid(&init_user_ns, inode->i_uid),
 819                      from_kgid(&init_user_ns, inode->i_gid));
 820         }
 821
 822         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 823             (issued & CEPH_CAP_LINK_EXCL) == 0)
 824                 set_nlink(inode, le32_to_cpu(info->nlink));
 825
 826         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 827                 /* be careful with mtime, atime, size */
 828                 ceph_decode_timespec64(&atime, &info->atime);
 829                 ceph_decode_timespec64(&mtime, &info->mtime);
 830                 ceph_decode_timespec64(&ctime, &info->ctime);
 831                 ceph_fill_file_time(inode, issued,
 832                                 le32_to_cpu(info->time_warp_seq),
 833                                 &ctime, &mtime, &atime);
 834         }
 835
 836         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 837                 ci->i_files = le64_to_cpu(info->files);
 838                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 839         }
 840
 841         if (new_version ||
 842             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 843                 s64 old_pool = ci->i_layout.pool_id;
 844                 struct ceph_string *old_ns;
 845
 846                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 847                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 848                                         lockdep_is_held(&ci->i_ceph_lock));
 849                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 850
 851                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 852                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 853
 854                 pool_ns = old_ns;
 855
 856                 queue_trunc = ceph_fill_file_size(inode, issued,
 857                                         le32_to_cpu(info->truncate_seq),
 858                                         le64_to_cpu(info->truncate_size),
 859                                         le64_to_cpu(info->size));
 860                 /* only update max_size on auth cap */
 861                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 862                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 863                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 864                                         le64_to_cpu(info->max_size));
 865                         ci->i_max_size = le64_to_cpu(info->max_size);
 866                 }
 867         }
 868
 869         /* layout and rstat are not tracked by capability, update them if
 870          * the inode info is from auth mds */
 871         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 872                 if (S_ISDIR(inode->i_mode)) {
 873                         ci->i_dir_layout = iinfo->dir_layout;
 874                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 875                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 876                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 877                         ci->i_dir_pin = iinfo->dir_pin;
 878                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 879                 }
 880         }
 881
 882         /* xattrs */
 883         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 884         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 885             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 886                 if (ci->i_xattrs.blob)
 887                         ceph_buffer_put(ci->i_xattrs.blob);
 888                 ci->i_xattrs.blob = xattr_blob;
 889                 if (xattr_blob)
 890                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 891                                iinfo->xattr_data, iinfo->xattr_len);
 892                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 893                 ceph_forget_all_cached_acls(inode);
 894                 xattr_blob = NULL;
 895         }
 896
 897         /* finally update i_version */
 898         if (le64_to_cpu(info->version) > ci->i_version)
 899                 ci->i_version = le64_to_cpu(info->version);
 900
 901         inode->i_mapping->a_ops = &ceph_aops;
 902
 903         switch (inode->i_mode & S_IFMT) {
 904         case S_IFIFO:
 905         case S_IFBLK:
 906         case S_IFCHR:
 907         case S_IFSOCK:
 908                 inode->i_blkbits = PAGE_SHIFT;
 909                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 910                 inode->i_op = &ceph_file_iops;
 911                 break;
 912         case S_IFREG:
 913                 inode->i_op = &ceph_file_iops;
 914                 inode->i_fop = &ceph_file_fops;
 915                 break;
 916         case S_IFLNK:
 917                 inode->i_op = &ceph_symlink_iops;
 918                 if (!ci->i_symlink) {
 919                         u32 symlen = iinfo->symlink_len;
 920                         char *sym;
 921
 922                         spin_unlock(&ci->i_ceph_lock);
 923
 924                         if (symlen != i_size_read(inode)) {
 925                                 pr_err("fill_inode %llx.%llx BAD symlink "
 926                                         "size %lld\n", ceph_vinop(inode),
 927                                         i_size_read(inode));
 928                                 i_size_write(inode, symlen);
 929                                 inode->i_blocks = calc_inode_blocks(symlen);
 930                         }
 931
 932                         err = -ENOMEM;
 933                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 934                         if (!sym)
 935                                 goto out;
 936
 937                         spin_lock(&ci->i_ceph_lock);
 938                         if (!ci->i_symlink)
 939                                 ci->i_symlink = sym;
 940                         else
 941                                 kfree(sym); /* lost a race */
 942                 }
 943                 inode->i_link = ci->i_symlink;
 944                 break;
 945         case S_IFDIR:
 946                 inode->i_op = &ceph_dir_iops;
 947                 inode->i_fop = &ceph_dir_fops;
 948                 break;
 949         default:
 950                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 951                        ceph_vinop(inode), inode->i_mode);
 952         }
 953
 954         /* were we issued a capability? */
 955         if (info_caps) {
 956                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 957                         ceph_add_cap(inode, session,
 958                                      le64_to_cpu(info->cap.cap_id),
 959                                      cap_fmode, info_caps,
 960                                      le32_to_cpu(info->cap.wanted),
 961                                      le32_to_cpu(info->cap.seq),
 962                                      le32_to_cpu(info->cap.mseq),
 963                                      le64_to_cpu(info->cap.realm),
 964                                      info->cap.flags, &new_cap);
 965
 966                         /* set dir completion flag? */
 967                         if (S_ISDIR(inode->i_mode) &&
 968                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 969                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 970                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 971                             !__ceph_dir_is_complete(ci)) {
 972                                 dout(" marking %p complete (empty)\n", inode);
 973                                 i_size_write(inode, 0);
 974                                 __ceph_dir_set_complete(ci,
 975                                         atomic64_read(&ci->i_release_count),
 976                                         atomic64_read(&ci->i_ordered_count));
 977                         }
 978
 979                         wake = true;
 980                 } else {
 981                         dout(" %p got snap_caps %s\n", inode,
 982                              ceph_cap_string(info_caps));
 983                         ci->i_snap_caps |= info_caps;
 984                         if (cap_fmode >= 0)
 985                                 __ceph_get_fmode(ci, cap_fmode);
 986                 }
 987         } else if (cap_fmode >= 0) {
 988                 pr_warn("mds issued no caps on %llx.%llx\n",
 989                            ceph_vinop(inode));
 990                 __ceph_get_fmode(ci, cap_fmode);
 991         }
 992
 993         if (iinfo->inline_version > 0 &&
 994             iinfo->inline_version >= ci->i_inline_version) {
 995                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 996                 ci->i_inline_version = iinfo->inline_version;
 997                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 998                     (locked_page || (info_caps & cache_caps)))
 999                         fill_inline = true;
1000         }
1001
1002         spin_unlock(&ci->i_ceph_lock);
1003
1004         if (fill_inline)
1005                 ceph_fill_inline_data(inode, locked_page,
1006                                       iinfo->inline_data, iinfo->inline_len);
1007
1008         if (wake)
1009                 wake_up_all(&ci->i_cap_wq);
1010
1011         /* queue truncate if we saw i_size decrease */
1012         if (queue_trunc)
1013                 ceph_queue_vmtruncate(inode);
1014
1015         /* populate frag tree */
1016         if (S_ISDIR(inode->i_mode))
1017                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1018
1019         /* update delegation info? */
1020         if (dirinfo)
1021                 ceph_fill_dirfrag(inode, dirinfo);
1022
1023         err = 0;
1024 out:
1025         if (new_cap)
1026                 ceph_put_cap(mdsc, new_cap);
1027         if (xattr_blob)
1028                 ceph_buffer_put(xattr_blob);
1029         ceph_put_string(pool_ns);
1030         return err;
1031 }
1032
1033 /*
1034  * caller should hold session s_mutex.
1035  */
1036 static void update_dentry_lease(struct dentry *dentry,
1037                                 struct ceph_mds_reply_lease *lease,
1038                                 struct ceph_mds_session *session,
1039                                 unsigned long from_time,
1040                                 struct ceph_vino *tgt_vino,
1041                                 struct ceph_vino *dir_vino)
1042 {
1043         struct ceph_dentry_info *di = ceph_dentry(dentry);
1044         long unsigned duration = le32_to_cpu(lease->duration_ms);
1045         long unsigned ttl = from_time + (duration * HZ) / 1000;
1046         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1047         struct inode *dir;
1048         struct ceph_mds_session *old_lease_session = NULL;
1049
1050         /*
1051          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1052          * we expect a negative dentry.
1053          */
1054         if (!tgt_vino && d_really_is_positive(dentry))
1055                 return;
1056
1057         if (tgt_vino && (d_really_is_negative(dentry) ||
1058                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1059                 return;
1060
1061         spin_lock(&dentry->d_lock);
1062         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1063              dentry, duration, ttl);
1064
1065         dir = d_inode(dentry->d_parent);
1066
1067         /* make sure parent matches dir_vino */
1068         if (!ceph_ino_compare(dir, dir_vino))
1069                 goto out_unlock;
1070
1071         /* only track leases on regular dentries */
1072         if (ceph_snap(dir) != CEPH_NOSNAP)
1073                 goto out_unlock;
1074
1075         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1076         if (duration == 0) {
1077                 __ceph_dentry_dir_lease_touch(di);
1078                 goto out_unlock;
1079         }
1080
1081         if (di->lease_gen == session->s_cap_gen &&
1082             time_before(ttl, di->time))
1083                 goto out_unlock;  /* we already have a newer lease. */
1084
1085         if (di->lease_session && di->lease_session != session) {
1086                 old_lease_session = di->lease_session;
1087                 di->lease_session = NULL;
1088         }
1089
1090         if (!di->lease_session)
1091                 di->lease_session = ceph_get_mds_session(session);
1092         di->lease_gen = session->s_cap_gen;
1093         di->lease_seq = le32_to_cpu(lease->seq);
1094         di->lease_renew_after = half_ttl;
1095         di->lease_renew_from = 0;
1096         di->time = ttl;
1097
1098         __ceph_dentry_lease_touch(di);
1099 out_unlock:
1100         spin_unlock(&dentry->d_lock);
1101         if (old_lease_session)
1102                 ceph_put_mds_session(old_lease_session);
1103 }
1104
1105 /*
1106  * splice a dentry to an inode.
1107  * caller must hold directory i_mutex for this to be safe.
1108  */
1109 static int splice_dentry(struct dentry **pdn, struct inode *in)
1110 {
1111         struct dentry *dn = *pdn;
1112         struct dentry *realdn;
1113
1114         BUG_ON(d_inode(dn));
1115
1116         if (S_ISDIR(in->i_mode)) {
1117                 /* If inode is directory, d_splice_alias() below will remove
1118                  * 'realdn' from its origin parent. We need to ensure that
1119                  * origin parent's readdir cache will not reference 'realdn'
1120                  */
1121                 realdn = d_find_any_alias(in);
1122                 if (realdn) {
1123                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1124                         spin_lock(&realdn->d_lock);
1125
1126                         realdn->d_op->d_prune(realdn);
1127
1128                         di->time = jiffies;
1129                         di->lease_shared_gen = 0;
1130                         di->offset = 0;
1131
1132                         spin_unlock(&realdn->d_lock);
1133                         dput(realdn);
1134                 }
1135         }
1136
1137         /* dn must be unhashed */
1138         if (!d_unhashed(dn))
1139                 d_drop(dn);
1140         realdn = d_splice_alias(in, dn);
1141         if (IS_ERR(realdn)) {
1142                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1143                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1144                 return PTR_ERR(realdn);
1145         }
1146
1147         if (realdn) {
1148                 dout("dn %p (%d) spliced with %p (%d) "
1149                      "inode %p ino %llx.%llx\n",
1150                      dn, d_count(dn),
1151                      realdn, d_count(realdn),
1152                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1153                 dput(dn);
1154                 *pdn = realdn;
1155         } else {
1156                 BUG_ON(!ceph_dentry(dn));
1157                 dout("dn %p attached to %p ino %llx.%llx\n",
1158                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1159         }
1160         return 0;
1161 }
1162
1163 static int d_name_cmp(struct dentry *dentry, const char *name, size_t len)
1164 {
1165         int ret;
1166
1167         /* take d_lock to ensure dentry->d_name stability */
1168         spin_lock(&dentry->d_lock);
1169         ret = dentry->d_name.len - len;
1170         if (!ret)
1171                 ret = memcmp(dentry->d_name.name, name, len);
1172         spin_unlock(&dentry->d_lock);
1173         return ret;
1174 }
1175
1176 /*
1177  * Incorporate results into the local cache.  This is either just
1178  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1179  * after a lookup).
1180  *
1181  * A reply may contain
1182  *         a directory inode along with a dentry.
1183  *  and/or a target inode
1184  *
1185  * Called with snap_rwsem (read).
1186  */
1187 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1188 {
1189         struct ceph_mds_session *session = req->r_session;
1190         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1191         struct inode *in = NULL;
1192         struct ceph_vino tvino, dvino;
1193         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1194         int err = 0;
1195
1196         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1197              rinfo->head->is_dentry, rinfo->head->is_target);
1198
1199         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1200                 dout("fill_trace reply is empty!\n");
1201                 if (rinfo->head->result == 0 && req->r_parent)
1202                         ceph_invalidate_dir_request(req);
1203                 return 0;
1204         }
1205
1206         if (rinfo->head->is_dentry) {
1207                 struct inode *dir = req->r_parent;
1208
1209                 if (dir) {
1210                         err = fill_inode(dir, NULL,
1211                                          &rinfo->diri, rinfo->dirfrag,
1212                                          session, req->r_request_started, -1,
1213                                          &req->r_caps_reservation);
1214                         if (err < 0)
1215                                 goto done;
1216                 } else {
1217                         WARN_ON_ONCE(1);
1218                 }
1219
1220                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1221                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1222                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1223                         struct qstr dname;
1224                         struct dentry *dn, *parent;
1225
1226                         BUG_ON(!rinfo->head->is_target);
1227                         BUG_ON(req->r_dentry);
1228
1229                         parent = d_find_any_alias(dir);
1230                         BUG_ON(!parent);
1231
1232                         dname.name = rinfo->dname;
1233                         dname.len = rinfo->dname_len;
1234                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1235                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1236                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1237 retry_lookup:
1238                         dn = d_lookup(parent, &dname);
1239                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1240                              parent, dname.len, dname.name, dn);
1241
1242                         if (!dn) {
1243                                 dn = d_alloc(parent, &dname);
1244                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1245                                      dname.len, dname.name, dn);
1246                                 if (!dn) {
1247                                         dput(parent);
1248                                         err = -ENOMEM;
1249                                         goto done;
1250                                 }
1251                                 err = 0;
1252                         } else if (d_really_is_positive(dn) &&
1253                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1254                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1255                                 dout(" dn %p points to wrong inode %p\n",
1256                                      dn, d_inode(dn));
1257                                 ceph_dir_clear_ordered(dir);
1258                                 d_delete(dn);
1259                                 dput(dn);
1260                                 goto retry_lookup;
1261                         }
1262
1263                         req->r_dentry = dn;
1264                         dput(parent);
1265                 }
1266         }
1267
1268         if (rinfo->head->is_target) {
1269                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1270                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1271
1272                 in = ceph_get_inode(sb, tvino);
1273                 if (IS_ERR(in)) {
1274                         err = PTR_ERR(in);
1275                         goto done;
1276                 }
1277                 req->r_target_inode = in;
1278
1279                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1280                                 session, req->r_request_started,
1281                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1282                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1283                                 &req->r_caps_reservation);
1284                 if (err < 0) {
1285                         pr_err("fill_inode badness %p %llx.%llx\n",
1286                                 in, ceph_vinop(in));
1287                         goto done;
1288                 }
1289         }
1290
1291         /*
1292          * ignore null lease/binding on snapdir ENOENT, or else we
1293          * will have trouble splicing in the virtual snapdir later
1294          */
1295         if (rinfo->head->is_dentry &&
1296             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1297             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1298             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1299                                                fsc->mount_options->snapdir_name,
1300                                                req->r_dentry->d_name.len))) {
1301                 /*
1302                  * lookup link rename   : null -> possibly existing inode
1303                  * mknod symlink mkdir  : null -> new inode
1304                  * unlink               : linked -> null
1305                  */
1306                 struct inode *dir = req->r_parent;
1307                 struct dentry *dn = req->r_dentry;
1308                 bool have_dir_cap, have_lease;
1309
1310                 BUG_ON(!dn);
1311                 BUG_ON(!dir);
1312                 BUG_ON(d_inode(dn->d_parent) != dir);
1313
1314                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1315                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1316
1317                 BUG_ON(ceph_ino(dir) != dvino.ino);
1318                 BUG_ON(ceph_snap(dir) != dvino.snap);
1319
1320                 /* do we have a lease on the whole dir? */
1321                 have_dir_cap =
1322                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1323                          CEPH_CAP_FILE_SHARED);
1324
1325                 /* do we have a dn lease? */
1326                 have_lease = have_dir_cap ||
1327                         le32_to_cpu(rinfo->dlease->duration_ms);
1328                 if (!have_lease)
1329                         dout("fill_trace  no dentry lease or dir cap\n");
1330
1331                 /* rename? */
1332                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1333                         struct inode *olddir = req->r_old_dentry_dir;
1334                         BUG_ON(!olddir);
1335
1336                         dout(" src %p '%pd' dst %p '%pd'\n",
1337                              req->r_old_dentry,
1338                              req->r_old_dentry,
1339                              dn, dn);
1340                         dout("fill_trace doing d_move %p -> %p\n",
1341                              req->r_old_dentry, dn);
1342
1343                         /* d_move screws up sibling dentries' offsets */
1344                         ceph_dir_clear_ordered(dir);
1345                         ceph_dir_clear_ordered(olddir);
1346
1347                         d_move(req->r_old_dentry, dn);
1348                         dout(" src %p '%pd' dst %p '%pd'\n",
1349                              req->r_old_dentry,
1350                              req->r_old_dentry,
1351                              dn, dn);
1352
1353                         /* ensure target dentry is invalidated, despite
1354                            rehashing bug in vfs_rename_dir */
1355                         ceph_invalidate_dentry_lease(dn);
1356
1357                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1358                              ceph_dentry(req->r_old_dentry)->offset);
1359
1360                         /* swap r_dentry and r_old_dentry in case that
1361                          * splice_dentry() gets called later. This is safe
1362                          * because no other place will use them */
1363                         req->r_dentry = req->r_old_dentry;
1364                         req->r_old_dentry = dn;
1365                         dn = req->r_dentry;
1366                 }
1367
1368                 /* null dentry? */
1369                 if (!rinfo->head->is_target) {
1370                         dout("fill_trace null dentry\n");
1371                         if (d_really_is_positive(dn)) {
1372                                 dout("d_delete %p\n", dn);
1373                                 ceph_dir_clear_ordered(dir);
1374                                 d_delete(dn);
1375                         } else if (have_lease) {
1376                                 if (d_unhashed(dn))
1377                                         d_add(dn, NULL);
1378                                 update_dentry_lease(dn, rinfo->dlease,
1379                                                     session,
1380                                                     req->r_request_started,
1381                                                     NULL, &dvino);
1382                         }
1383                         goto done;
1384                 }
1385
1386                 /* attach proper inode */
1387                 if (d_really_is_negative(dn)) {
1388                         ceph_dir_clear_ordered(dir);
1389                         ihold(in);
1390                         err = splice_dentry(&req->r_dentry, in);
1391                         if (err < 0)
1392                                 goto done;
1393                         dn = req->r_dentry;  /* may have spliced */
1394                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1395                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1396                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1397                              ceph_vinop(in));
1398                         d_invalidate(dn);
1399                         have_lease = false;
1400                 }
1401
1402                 if (have_lease) {
1403                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1404                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1405                         update_dentry_lease(dn, rinfo->dlease, session,
1406                                             req->r_request_started,
1407                                             &tvino, &dvino);
1408                 }
1409                 dout(" final dn %p\n", dn);
1410         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1411                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1412                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1413                 struct inode *dir = req->r_parent;
1414
1415                 /* fill out a snapdir LOOKUPSNAP dentry */
1416                 BUG_ON(!dir);
1417                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1418                 BUG_ON(!req->r_dentry);
1419                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1420                 ceph_dir_clear_ordered(dir);
1421                 ihold(in);
1422                 err = splice_dentry(&req->r_dentry, in);
1423                 if (err < 0)
1424                         goto done;
1425         } else if (rinfo->head->is_dentry &&
1426                    !d_name_cmp(req->r_dentry, rinfo->dname, rinfo->dname_len)) {
1427                 struct ceph_vino *ptvino = NULL;
1428
1429                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1430                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1431                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1432                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1433
1434                         if (rinfo->head->is_target) {
1435                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1436                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1437                                 ptvino = &tvino;
1438                         }
1439
1440                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1441                                 session, req->r_request_started, ptvino,
1442                                 &dvino);
1443                 } else {
1444                         dout("%s: no dentry lease or dir cap\n", __func__);
1445                 }
1446         }
1447 done:
1448         dout("fill_trace done err=%d\n", err);
1449         return err;
1450 }
1451
1452 /*
1453  * Prepopulate our cache with readdir results, leases, etc.
1454  */
1455 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1456                                            struct ceph_mds_session *session)
1457 {
1458         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1459         int i, err = 0;
1460
1461         for (i = 0; i < rinfo->dir_nr; i++) {
1462                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1463                 struct ceph_vino vino;
1464                 struct inode *in;
1465                 int rc;
1466
1467                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1468                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1469
1470                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1471                 if (IS_ERR(in)) {
1472                         err = PTR_ERR(in);
1473                         dout("new_inode badness got %d\n", err);
1474                         continue;
1475                 }
1476                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1477                                 req->r_request_started, -1,
1478                                 &req->r_caps_reservation);
1479                 if (rc < 0) {
1480                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1481                         err = rc;
1482                 }
1483                 iput(in);
1484         }
1485
1486         return err;
1487 }
1488
1489 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1490 {
1491         if (ctl->page) {
1492                 kunmap(ctl->page);
1493                 put_page(ctl->page);
1494                 ctl->page = NULL;
1495         }
1496 }
1497
1498 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1499                               struct ceph_readdir_cache_control *ctl,
1500                               struct ceph_mds_request *req)
1501 {
1502         struct ceph_inode_info *ci = ceph_inode(dir);
1503         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1504         unsigned idx = ctl->index % nsize;
1505         pgoff_t pgoff = ctl->index / nsize;
1506
1507         if (!ctl->page || pgoff != page_index(ctl->page)) {
1508                 ceph_readdir_cache_release(ctl);
1509                 if (idx == 0)
1510                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1511                 else
1512                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1513                 if (!ctl->page) {
1514                         ctl->index = -1;
1515                         return idx == 0 ? -ENOMEM : 0;
1516                 }
1517                 /* reading/filling the cache are serialized by
1518                  * i_mutex, no need to use page lock */
1519                 unlock_page(ctl->page);
1520                 ctl->dentries = kmap(ctl->page);
1521                 if (idx == 0)
1522                         memset(ctl->dentries, 0, PAGE_SIZE);
1523         }
1524
1525         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1526             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1527                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1528                 ctl->dentries[idx] = dn;
1529                 ctl->index++;
1530         } else {
1531                 dout("disable readdir cache\n");
1532                 ctl->index = -1;
1533         }
1534         return 0;
1535 }
1536
1537 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1538                              struct ceph_mds_session *session)
1539 {
1540         struct dentry *parent = req->r_dentry;
1541         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1542         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1543         struct qstr dname;
1544         struct dentry *dn;
1545         struct inode *in;
1546         int err = 0, skipped = 0, ret, i;
1547         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1548         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1549         u32 last_hash = 0;
1550         u32 fpos_offset;
1551         struct ceph_readdir_cache_control cache_ctl = {};
1552
1553         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1554                 return readdir_prepopulate_inodes_only(req, session);
1555
1556         if (rinfo->hash_order) {
1557                 if (req->r_path2) {
1558                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1559                                                   req->r_path2,
1560                                                   strlen(req->r_path2));
1561                         last_hash = ceph_frag_value(last_hash);
1562                 } else if (rinfo->offset_hash) {
1563                         /* mds understands offset_hash */
1564                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1565                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1566                 }
1567         }
1568
1569         if (rinfo->dir_dir &&
1570             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1571                 dout("readdir_prepopulate got new frag %x -> %x\n",
1572                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1573                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1574                 if (!rinfo->hash_order)
1575                         req->r_readdir_offset = 2;
1576         }
1577
1578         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1579                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1580                      rinfo->dir_nr, parent);
1581         } else {
1582                 dout("readdir_prepopulate %d items under dn %p\n",
1583                      rinfo->dir_nr, parent);
1584                 if (rinfo->dir_dir)
1585                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1586
1587                 if (ceph_frag_is_leftmost(frag) &&
1588                     req->r_readdir_offset == 2 &&
1589                     !(rinfo->hash_order && last_hash)) {
1590                         /* note dir version at start of readdir so we can
1591                          * tell if any dentries get dropped */
1592                         req->r_dir_release_cnt =
1593                                 atomic64_read(&ci->i_release_count);
1594                         req->r_dir_ordered_cnt =
1595                                 atomic64_read(&ci->i_ordered_count);
1596                         req->r_readdir_cache_idx = 0;
1597                 }
1598         }
1599
1600         cache_ctl.index = req->r_readdir_cache_idx;
1601         fpos_offset = req->r_readdir_offset;
1602
1603         /* FIXME: release caps/leases if error occurs */
1604         for (i = 0; i < rinfo->dir_nr; i++) {
1605                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1606                 struct ceph_vino tvino, dvino;
1607
1608                 dname.name = rde->name;
1609                 dname.len = rde->name_len;
1610                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1611
1612                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1613                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1614
1615                 if (rinfo->hash_order) {
1616                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1617                                                  rde->name, rde->name_len);
1618                         hash = ceph_frag_value(hash);
1619                         if (hash != last_hash)
1620                                 fpos_offset = 2;
1621                         last_hash = hash;
1622                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1623                 } else {
1624                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1625                 }
1626
1627 retry_lookup:
1628                 dn = d_lookup(parent, &dname);
1629                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1630                      parent, dname.len, dname.name, dn);
1631
1632                 if (!dn) {
1633                         dn = d_alloc(parent, &dname);
1634                         dout("d_alloc %p '%.*s' = %p\n", parent,
1635                              dname.len, dname.name, dn);
1636                         if (!dn) {
1637                                 dout("d_alloc badness\n");
1638                                 err = -ENOMEM;
1639                                 goto out;
1640                         }
1641                 } else if (d_really_is_positive(dn) &&
1642                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1643                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1644                         struct ceph_dentry_info *di = ceph_dentry(dn);
1645                         dout(" dn %p points to wrong inode %p\n",
1646                              dn, d_inode(dn));
1647
1648                         spin_lock(&dn->d_lock);
1649                         if (di->offset > 0 &&
1650                             di->lease_shared_gen ==
1651                             atomic_read(&ci->i_shared_gen)) {
1652                                 __ceph_dir_clear_ordered(ci);
1653                                 di->offset = 0;
1654                         }
1655                         spin_unlock(&dn->d_lock);
1656
1657                         d_delete(dn);
1658                         dput(dn);
1659                         goto retry_lookup;
1660                 }
1661
1662                 /* inode */
1663                 if (d_really_is_positive(dn)) {
1664                         in = d_inode(dn);
1665                 } else {
1666                         in = ceph_get_inode(parent->d_sb, tvino);
1667                         if (IS_ERR(in)) {
1668                                 dout("new_inode badness\n");
1669                                 d_drop(dn);
1670                                 dput(dn);
1671                                 err = PTR_ERR(in);
1672                                 goto out;
1673                         }
1674                 }
1675
1676                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1677                                  req->r_request_started, -1,
1678                                  &req->r_caps_reservation);
1679                 if (ret < 0) {
1680                         pr_err("fill_inode badness on %p\n", in);
1681                         if (d_really_is_negative(dn))
1682                                 iput(in);
1683                         d_drop(dn);
1684                         err = ret;
1685                         goto next_item;
1686                 }
1687
1688                 if (d_really_is_negative(dn)) {
1689                         if (ceph_security_xattr_deadlock(in)) {
1690                                 dout(" skip splicing dn %p to inode %p"
1691                                      " (security xattr deadlock)\n", dn, in);
1692                                 iput(in);
1693                                 skipped++;
1694                                 goto next_item;
1695                         }
1696
1697                         err = splice_dentry(&dn, in);
1698                         if (err < 0)
1699                                 goto next_item;
1700                 }
1701
1702                 ceph_dentry(dn)->offset = rde->offset;
1703
1704                 dvino = ceph_vino(d_inode(parent));
1705                 update_dentry_lease(dn, rde->lease, req->r_session,
1706                                     req->r_request_started, &tvino, &dvino);
1707
1708                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1709                         ret = fill_readdir_cache(d_inode(parent), dn,
1710                                                  &cache_ctl, req);
1711                         if (ret < 0)
1712                                 err = ret;
1713                 }
1714 next_item:
1715                 dput(dn);
1716         }
1717 out:
1718         if (err == 0 && skipped == 0) {
1719                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1720                 req->r_readdir_cache_idx = cache_ctl.index;
1721         }
1722         ceph_readdir_cache_release(&cache_ctl);
1723         dout("readdir_prepopulate done\n");
1724         return err;
1725 }
1726
1727 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1728 {
1729         struct ceph_inode_info *ci = ceph_inode(inode);
1730         bool ret;
1731
1732         spin_lock(&ci->i_ceph_lock);
1733         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1734         i_size_write(inode, size);
1735         inode->i_blocks = calc_inode_blocks(size);
1736
1737         ret = __ceph_should_report_size(ci);
1738
1739         spin_unlock(&ci->i_ceph_lock);
1740         return ret;
1741 }
1742
1743 /*
1744  * Write back inode data in a worker thread.  (This can't be done
1745  * in the message handler context.)
1746  */
1747 void ceph_queue_writeback(struct inode *inode)
1748 {
1749         ihold(inode);
1750         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1751                        &ceph_inode(inode)->i_wb_work)) {
1752                 dout("ceph_queue_writeback %p\n", inode);
1753         } else {
1754                 dout("ceph_queue_writeback %p failed\n", inode);
1755                 iput(inode);
1756         }
1757 }
1758
1759 static void ceph_writeback_work(struct work_struct *work)
1760 {
1761         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1762                                                   i_wb_work);
1763         struct inode *inode = &ci->vfs_inode;
1764
1765         dout("writeback %p\n", inode);
1766         filemap_fdatawrite(&inode->i_data);
1767         iput(inode);
1768 }
1769
1770 /*
1771  * queue an async invalidation
1772  */
1773 void ceph_queue_invalidate(struct inode *inode)
1774 {
1775         ihold(inode);
1776         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1777                        &ceph_inode(inode)->i_pg_inv_work)) {
1778                 dout("ceph_queue_invalidate %p\n", inode);
1779         } else {
1780                 dout("ceph_queue_invalidate %p failed\n", inode);
1781                 iput(inode);
1782         }
1783 }
1784
1785 /*
1786  * Invalidate inode pages in a worker thread.  (This can't be done
1787  * in the message handler context.)
1788  */
1789 static void ceph_invalidate_work(struct work_struct *work)
1790 {
1791         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1792                                                   i_pg_inv_work);
1793         struct inode *inode = &ci->vfs_inode;
1794         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1795         u32 orig_gen;
1796         int check = 0;
1797
1798         mutex_lock(&ci->i_truncate_mutex);
1799
1800         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1801                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1802                                     inode, ceph_ino(inode));
1803                 mapping_set_error(inode->i_mapping, -EIO);
1804                 truncate_pagecache(inode, 0);
1805                 mutex_unlock(&ci->i_truncate_mutex);
1806                 goto out;
1807         }
1808
1809         spin_lock(&ci->i_ceph_lock);
1810         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1811              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1812         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1813                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1814                         check = 1;
1815                 spin_unlock(&ci->i_ceph_lock);
1816                 mutex_unlock(&ci->i_truncate_mutex);
1817                 goto out;
1818         }
1819         orig_gen = ci->i_rdcache_gen;
1820         spin_unlock(&ci->i_ceph_lock);
1821
1822         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1823                 pr_err("invalidate_pages %p fails\n", inode);
1824         }
1825
1826         spin_lock(&ci->i_ceph_lock);
1827         if (orig_gen == ci->i_rdcache_gen &&
1828             orig_gen == ci->i_rdcache_revoking) {
1829                 dout("invalidate_pages %p gen %d successful\n", inode,
1830                      ci->i_rdcache_gen);
1831                 ci->i_rdcache_revoking--;
1832                 check = 1;
1833         } else {
1834                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1835                      inode, orig_gen, ci->i_rdcache_gen,
1836                      ci->i_rdcache_revoking);
1837                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1838                         check = 1;
1839         }
1840         spin_unlock(&ci->i_ceph_lock);
1841         mutex_unlock(&ci->i_truncate_mutex);
1842 out:
1843         if (check)
1844                 ceph_check_caps(ci, 0, NULL);
1845         iput(inode);
1846 }
1847
1848
1849 /*
1850  * called by trunc_wq;
1851  *
1852  * We also truncate in a separate thread as well.
1853  */
1854 static void ceph_vmtruncate_work(struct work_struct *work)
1855 {
1856         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1857                                                   i_vmtruncate_work);
1858         struct inode *inode = &ci->vfs_inode;
1859
1860         dout("vmtruncate_work %p\n", inode);
1861         __ceph_do_pending_vmtruncate(inode);
1862         iput(inode);
1863 }
1864
1865 /*
1866  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1867  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1868  */
1869 void ceph_queue_vmtruncate(struct inode *inode)
1870 {
1871         struct ceph_inode_info *ci = ceph_inode(inode);
1872
1873         ihold(inode);
1874
1875         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1876                        &ci->i_vmtruncate_work)) {
1877                 dout("ceph_queue_vmtruncate %p\n", inode);
1878         } else {
1879                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1880                      inode, ci->i_truncate_pending);
1881                 iput(inode);
1882         }
1883 }
1884
1885 /*
1886  * Make sure any pending truncation is applied before doing anything
1887  * that may depend on it.
1888  */
1889 void __ceph_do_pending_vmtruncate(struct inode *inode)
1890 {
1891         struct ceph_inode_info *ci = ceph_inode(inode);
1892         u64 to;
1893         int wrbuffer_refs, finish = 0;
1894
1895         mutex_lock(&ci->i_truncate_mutex);
1896 retry:
1897         spin_lock(&ci->i_ceph_lock);
1898         if (ci->i_truncate_pending == 0) {
1899                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1900                 spin_unlock(&ci->i_ceph_lock);
1901                 mutex_unlock(&ci->i_truncate_mutex);
1902                 return;
1903         }
1904
1905         /*
1906          * make sure any dirty snapped pages are flushed before we
1907          * possibly truncate them.. so write AND block!
1908          */
1909         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1910                 spin_unlock(&ci->i_ceph_lock);
1911                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1912                      inode);
1913                 filemap_write_and_wait_range(&inode->i_data, 0,
1914                                              inode->i_sb->s_maxbytes);
1915                 goto retry;
1916         }
1917
1918         /* there should be no reader or writer */
1919         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1920
1921         to = ci->i_truncate_size;
1922         wrbuffer_refs = ci->i_wrbuffer_ref;
1923         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1924              ci->i_truncate_pending, to);
1925         spin_unlock(&ci->i_ceph_lock);
1926
1927         truncate_pagecache(inode, to);
1928
1929         spin_lock(&ci->i_ceph_lock);
1930         if (to == ci->i_truncate_size) {
1931                 ci->i_truncate_pending = 0;
1932                 finish = 1;
1933         }
1934         spin_unlock(&ci->i_ceph_lock);
1935         if (!finish)
1936                 goto retry;
1937
1938         mutex_unlock(&ci->i_truncate_mutex);
1939
1940         if (wrbuffer_refs == 0)
1941                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1942
1943         wake_up_all(&ci->i_cap_wq);
1944 }
1945
1946 /*
1947  * symlinks
1948  */
1949 static const struct inode_operations ceph_symlink_iops = {
1950         .get_link = simple_get_link,
1951         .setattr = ceph_setattr,
1952         .getattr = ceph_getattr,
1953         .listxattr = ceph_listxattr,
1954 };
1955
1956 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1957 {
1958         struct ceph_inode_info *ci = ceph_inode(inode);
1959         const unsigned int ia_valid = attr->ia_valid;
1960         struct ceph_mds_request *req;
1961         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1962         struct ceph_cap_flush *prealloc_cf;
1963         int issued;
1964         int release = 0, dirtied = 0;
1965         int mask = 0;
1966         int err = 0;
1967         int inode_dirty_flags = 0;
1968         bool lock_snap_rwsem = false;
1969
1970         prealloc_cf = ceph_alloc_cap_flush();
1971         if (!prealloc_cf)
1972                 return -ENOMEM;
1973
1974         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1975                                        USE_AUTH_MDS);
1976         if (IS_ERR(req)) {
1977                 ceph_free_cap_flush(prealloc_cf);
1978                 return PTR_ERR(req);
1979         }
1980
1981         spin_lock(&ci->i_ceph_lock);
1982         issued = __ceph_caps_issued(ci, NULL);
1983
1984         if (!ci->i_head_snapc &&
1985             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1986                 lock_snap_rwsem = true;
1987                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1988                         spin_unlock(&ci->i_ceph_lock);
1989                         down_read(&mdsc->snap_rwsem);
1990                         spin_lock(&ci->i_ceph_lock);
1991                         issued = __ceph_caps_issued(ci, NULL);
1992                 }
1993         }
1994
1995         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1996
1997         if (ia_valid & ATTR_UID) {
1998                 dout("setattr %p uid %d -> %d\n", inode,
1999                      from_kuid(&init_user_ns, inode->i_uid),
2000                      from_kuid(&init_user_ns, attr->ia_uid));
2001                 if (issued & CEPH_CAP_AUTH_EXCL) {
2002                         inode->i_uid = attr->ia_uid;
2003                         dirtied |= CEPH_CAP_AUTH_EXCL;
2004                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2005                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2006                         req->r_args.setattr.uid = cpu_to_le32(
2007                                 from_kuid(&init_user_ns, attr->ia_uid));
2008                         mask |= CEPH_SETATTR_UID;
2009                         release |= CEPH_CAP_AUTH_SHARED;
2010                 }
2011         }
2012         if (ia_valid & ATTR_GID) {
2013                 dout("setattr %p gid %d -> %d\n", inode,
2014                      from_kgid(&init_user_ns, inode->i_gid),
2015                      from_kgid(&init_user_ns, attr->ia_gid));
2016                 if (issued & CEPH_CAP_AUTH_EXCL) {
2017                         inode->i_gid = attr->ia_gid;
2018                         dirtied |= CEPH_CAP_AUTH_EXCL;
2019                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2020                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2021                         req->r_args.setattr.gid = cpu_to_le32(
2022                                 from_kgid(&init_user_ns, attr->ia_gid));
2023                         mask |= CEPH_SETATTR_GID;
2024                         release |= CEPH_CAP_AUTH_SHARED;
2025                 }
2026         }
2027         if (ia_valid & ATTR_MODE) {
2028                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2029                      attr->ia_mode);
2030                 if (issued & CEPH_CAP_AUTH_EXCL) {
2031                         inode->i_mode = attr->ia_mode;
2032                         dirtied |= CEPH_CAP_AUTH_EXCL;
2033                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2034                            attr->ia_mode != inode->i_mode) {
2035                         inode->i_mode = attr->ia_mode;
2036                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2037                         mask |= CEPH_SETATTR_MODE;
2038                         release |= CEPH_CAP_AUTH_SHARED;
2039                 }
2040         }
2041
2042         if (ia_valid & ATTR_ATIME) {
2043                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2044                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2045                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2046                 if (issued & CEPH_CAP_FILE_EXCL) {
2047                         ci->i_time_warp_seq++;
2048                         inode->i_atime = attr->ia_atime;
2049                         dirtied |= CEPH_CAP_FILE_EXCL;
2050                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2051                            timespec64_compare(&inode->i_atime,
2052                                             &attr->ia_atime) < 0) {
2053                         inode->i_atime = attr->ia_atime;
2054                         dirtied |= CEPH_CAP_FILE_WR;
2055                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2056                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2057                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2058                                                &attr->ia_atime);
2059                         mask |= CEPH_SETATTR_ATIME;
2060                         release |= CEPH_CAP_FILE_SHARED |
2061                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2062                 }
2063         }
2064         if (ia_valid & ATTR_MTIME) {
2065                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2066                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2067                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2068                 if (issued & CEPH_CAP_FILE_EXCL) {
2069                         ci->i_time_warp_seq++;
2070                         inode->i_mtime = attr->ia_mtime;
2071                         dirtied |= CEPH_CAP_FILE_EXCL;
2072                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2073                            timespec64_compare(&inode->i_mtime,
2074                                             &attr->ia_mtime) < 0) {
2075                         inode->i_mtime = attr->ia_mtime;
2076                         dirtied |= CEPH_CAP_FILE_WR;
2077                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2078                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2079                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2080                                                &attr->ia_mtime);
2081                         mask |= CEPH_SETATTR_MTIME;
2082                         release |= CEPH_CAP_FILE_SHARED |
2083                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2084                 }
2085         }
2086         if (ia_valid & ATTR_SIZE) {
2087                 dout("setattr %p size %lld -> %lld\n", inode,
2088                      inode->i_size, attr->ia_size);
2089                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2090                     attr->ia_size > inode->i_size) {
2091                         i_size_write(inode, attr->ia_size);
2092                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2093                         ci->i_reported_size = attr->ia_size;
2094                         dirtied |= CEPH_CAP_FILE_EXCL;
2095                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2096                            attr->ia_size != inode->i_size) {
2097                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2098                         req->r_args.setattr.old_size =
2099                                 cpu_to_le64(inode->i_size);
2100                         mask |= CEPH_SETATTR_SIZE;
2101                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2102                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2103                 }
2104         }
2105
2106         /* these do nothing */
2107         if (ia_valid & ATTR_CTIME) {
2108                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2109                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2110                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2111                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2112                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2113                      only ? "ctime only" : "ignored");
2114                 if (only) {
2115                         /*
2116                          * if kernel wants to dirty ctime but nothing else,
2117                          * we need to choose a cap to dirty under, or do
2118                          * a almost-no-op setattr
2119                          */
2120                         if (issued & CEPH_CAP_AUTH_EXCL)
2121                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2122                         else if (issued & CEPH_CAP_FILE_EXCL)
2123                                 dirtied |= CEPH_CAP_FILE_EXCL;
2124                         else if (issued & CEPH_CAP_XATTR_EXCL)
2125                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2126                         else
2127                                 mask |= CEPH_SETATTR_CTIME;
2128                 }
2129         }
2130         if (ia_valid & ATTR_FILE)
2131                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2132
2133         if (dirtied) {
2134                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2135                                                            &prealloc_cf);
2136                 inode->i_ctime = attr->ia_ctime;
2137         }
2138
2139         release &= issued;
2140         spin_unlock(&ci->i_ceph_lock);
2141         if (lock_snap_rwsem)
2142                 up_read(&mdsc->snap_rwsem);
2143
2144         if (inode_dirty_flags)
2145                 __mark_inode_dirty(inode, inode_dirty_flags);
2146
2147
2148         if (mask) {
2149                 req->r_inode = inode;
2150                 ihold(inode);
2151                 req->r_inode_drop = release;
2152                 req->r_args.setattr.mask = cpu_to_le32(mask);
2153                 req->r_num_caps = 1;
2154                 req->r_stamp = attr->ia_ctime;
2155                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2156         }
2157         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2158              ceph_cap_string(dirtied), mask);
2159
2160         ceph_mdsc_put_request(req);
2161         ceph_free_cap_flush(prealloc_cf);
2162
2163         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2164                 __ceph_do_pending_vmtruncate(inode);
2165
2166         return err;
2167 }
2168
2169 /*
2170  * setattr
2171  */
2172 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2173 {
2174         struct inode *inode = d_inode(dentry);
2175         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2176         int err;
2177
2178         if (ceph_snap(inode) != CEPH_NOSNAP)
2179                 return -EROFS;
2180
2181         err = setattr_prepare(dentry, attr);
2182         if (err != 0)
2183                 return err;
2184
2185         if ((attr->ia_valid & ATTR_SIZE) &&
2186             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2187                 return -EFBIG;
2188
2189         if ((attr->ia_valid & ATTR_SIZE) &&
2190             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2191                 return -EDQUOT;
2192
2193         err = __ceph_setattr(inode, attr);
2194
2195         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2196                 err = posix_acl_chmod(inode, attr->ia_mode);
2197
2198         return err;
2199 }
2200
2201 /*
2202  * Verify that we have a lease on the given mask.  If not,
2203  * do a getattr against an mds.
2204  */
2205 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2206                       int mask, bool force)
2207 {
2208         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2209         struct ceph_mds_client *mdsc = fsc->mdsc;
2210         struct ceph_mds_request *req;
2211         int mode;
2212         int err;
2213
2214         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2215                 dout("do_getattr inode %p SNAPDIR\n", inode);
2216                 return 0;
2217         }
2218
2219         dout("do_getattr inode %p mask %s mode 0%o\n",
2220              inode, ceph_cap_string(mask), inode->i_mode);
2221         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2222                 return 0;
2223
2224         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2225         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2226         if (IS_ERR(req))
2227                 return PTR_ERR(req);
2228         req->r_inode = inode;
2229         ihold(inode);
2230         req->r_num_caps = 1;
2231         req->r_args.getattr.mask = cpu_to_le32(mask);
2232         req->r_locked_page = locked_page;
2233         err = ceph_mdsc_do_request(mdsc, NULL, req);
2234         if (locked_page && err == 0) {
2235                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2236                 if (inline_version == 0) {
2237                         /* the reply is supposed to contain inline data */
2238                         err = -EINVAL;
2239                 } else if (inline_version == CEPH_INLINE_NONE) {
2240                         err = -ENODATA;
2241                 } else {
2242                         err = req->r_reply_info.targeti.inline_len;
2243                 }
2244         }
2245         ceph_mdsc_put_request(req);
2246         dout("do_getattr result=%d\n", err);
2247         return err;
2248 }
2249
2250
2251 /*
2252  * Check inode permissions.  We verify we have a valid value for
2253  * the AUTH cap, then call the generic handler.
2254  */
2255 int ceph_permission(struct inode *inode, int mask)
2256 {
2257         int err;
2258
2259         if (mask & MAY_NOT_BLOCK)
2260                 return -ECHILD;
2261
2262         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2263
2264         if (!err)
2265                 err = generic_permission(inode, mask);
2266         return err;
2267 }
2268
2269 /* Craft a mask of needed caps given a set of requested statx attrs. */
2270 static int statx_to_caps(u32 want)
2271 {
2272         int mask = 0;
2273
2274         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME))
2275                 mask |= CEPH_CAP_AUTH_SHARED;
2276
2277         if (want & (STATX_NLINK|STATX_CTIME))
2278                 mask |= CEPH_CAP_LINK_SHARED;
2279
2280         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2281                     STATX_BLOCKS))
2282                 mask |= CEPH_CAP_FILE_SHARED;
2283
2284         if (want & (STATX_CTIME))
2285                 mask |= CEPH_CAP_XATTR_SHARED;
2286
2287         return mask;
2288 }
2289
2290 /*
2291  * Get all the attributes. If we have sufficient caps for the requested attrs,
2292  * then we can avoid talking to the MDS at all.
2293  */
2294 int ceph_getattr(const struct path *path, struct kstat *stat,
2295                  u32 request_mask, unsigned int flags)
2296 {
2297         struct inode *inode = d_inode(path->dentry);
2298         struct ceph_inode_info *ci = ceph_inode(inode);
2299         int err = 0;
2300
2301         /* Skip the getattr altogether if we're asked not to sync */
2302         if (!(flags & AT_STATX_DONT_SYNC)) {
2303                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2304                                       flags & AT_STATX_FORCE_SYNC);
2305                 if (err)
2306                         return err;
2307         }
2308
2309         generic_fillattr(inode, stat);
2310         stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2311         if (ceph_snap(inode) == CEPH_NOSNAP)
2312                 stat->dev = inode->i_sb->s_dev;
2313         else
2314                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2315
2316         if (S_ISDIR(inode->i_mode)) {
2317                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2318                                         RBYTES))
2319                         stat->size = ci->i_rbytes;
2320                 else
2321                         stat->size = ci->i_files + ci->i_subdirs;
2322                 stat->blocks = 0;
2323                 stat->blksize = 65536;
2324                 /*
2325                  * Some applications rely on the number of st_nlink
2326                  * value on directories to be either 0 (if unlinked)
2327                  * or 2 + number of subdirectories.
2328                  */
2329                 if (stat->nlink == 1)
2330                         /* '.' + '..' + subdirs */
2331                         stat->nlink = 1 + 1 + ci->i_subdirs;
2332         }
2333
2334         /* Mask off any higher bits (e.g. btime) until we have support */
2335         stat->result_mask = request_mask & STATX_BASIC_STATS;
2336         return err;
2337 }