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 #include <linux/iversion.h>
  17
  18 #include "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21 #include <linux/ceph/decode.h>
  22
  23 /*
  24  * Ceph inode operations
  25  *
  26  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  27  * setattr, etc.), xattr helpers, and helpers for assimilating
  28  * metadata returned by the MDS into our cache.
  29  *
  30  * Also define helpers for doing asynchronous writeback, invalidation,
  31  * and truncation for the benefit of those who can't afford to block
  32  * (typically because they are in the message handler path).
  33  */
  34
  35 static const struct inode_operations ceph_symlink_iops;
  36
  37 static void ceph_inode_work(struct work_struct *work);
  38
  39 /*
  40  * find or create an inode, given the ceph ino number
  41  */
  42 static int ceph_set_ino_cb(struct inode *inode, void *data)
  43 {
  44         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  45         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  46         inode_set_iversion_raw(inode, 0);
  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_work, ceph_inode_work);
 513         ci->i_work_mask = 0;
 514         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 515
 516         ceph_fscache_inode_init(ci);
 517
 518         return &ci->vfs_inode;
 519 }
 520
 521 void ceph_free_inode(struct inode *inode)
 522 {
 523         struct ceph_inode_info *ci = ceph_inode(inode);
 524
 525         kfree(ci->i_symlink);
 526         kmem_cache_free(ceph_inode_cachep, ci);
 527 }
 528
 529 void ceph_evict_inode(struct inode *inode)
 530 {
 531         struct ceph_inode_info *ci = ceph_inode(inode);
 532         struct ceph_inode_frag *frag;
 533         struct rb_node *n;
 534
 535         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 536
 537         truncate_inode_pages_final(&inode->i_data);
 538         clear_inode(inode);
 539
 540         ceph_fscache_unregister_inode_cookie(ci);
 541
 542         __ceph_remove_caps(ci);
 543
 544         if (__ceph_has_any_quota(ci))
 545                 ceph_adjust_quota_realms_count(inode, false);
 546
 547         /*
 548          * we may still have a snap_realm reference if there are stray
 549          * caps in i_snap_caps.
 550          */
 551         if (ci->i_snap_realm) {
 552                 struct ceph_mds_client *mdsc =
 553                                         ceph_inode_to_client(inode)->mdsc;
 554                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 555                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 556                         dout(" dropping residual ref to snap realm %p\n",
 557                              realm);
 558                         spin_lock(&realm->inodes_with_caps_lock);
 559                         list_del_init(&ci->i_snap_realm_item);
 560                         ci->i_snap_realm = NULL;
 561                         if (realm->ino == ci->i_vino.ino)
 562                                 realm->inode = NULL;
 563                         spin_unlock(&realm->inodes_with_caps_lock);
 564                         ceph_put_snap_realm(mdsc, realm);
 565                 } else {
 566                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 567                         ci->i_snap_realm = NULL;
 568                 }
 569         }
 570
 571         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 572                 frag = rb_entry(n, struct ceph_inode_frag, node);
 573                 rb_erase(n, &ci->i_fragtree);
 574                 kfree(frag);
 575         }
 576         ci->i_fragtree_nsplits = 0;
 577
 578         __ceph_destroy_xattrs(ci);
 579         if (ci->i_xattrs.blob)
 580                 ceph_buffer_put(ci->i_xattrs.blob);
 581         if (ci->i_xattrs.prealloc_blob)
 582                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 583
 584         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 585 }
 586
 587 static inline blkcnt_t calc_inode_blocks(u64 size)
 588 {
 589         return (size + (1<<9) - 1) >> 9;
 590 }
 591
 592 /*
 593  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 594  * careful because either the client or MDS may have more up to date
 595  * info, depending on which capabilities are held, and whether
 596  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 597  * and size are monotonically increasing, except when utimes() or
 598  * truncate() increments the corresponding _seq values.)
 599  */
 600 int ceph_fill_file_size(struct inode *inode, int issued,
 601                         u32 truncate_seq, u64 truncate_size, u64 size)
 602 {
 603         struct ceph_inode_info *ci = ceph_inode(inode);
 604         int queue_trunc = 0;
 605
 606         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 607             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 608                 dout("size %lld -> %llu\n", inode->i_size, size);
 609                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 610                         pr_err("fill_file_size non-zero size for directory\n");
 611                         size = 0;
 612                 }
 613                 i_size_write(inode, size);
 614                 inode->i_blocks = calc_inode_blocks(size);
 615                 ci->i_reported_size = size;
 616                 if (truncate_seq != ci->i_truncate_seq) {
 617                         dout("truncate_seq %u -> %u\n",
 618                              ci->i_truncate_seq, truncate_seq);
 619                         ci->i_truncate_seq = truncate_seq;
 620
 621                         /* the MDS should have revoked these caps */
 622                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 623                                                CEPH_CAP_FILE_RD |
 624                                                CEPH_CAP_FILE_WR |
 625                                                CEPH_CAP_FILE_LAZYIO));
 626                         /*
 627                          * If we hold relevant caps, or in the case where we're
 628                          * not the only client referencing this file and we
 629                          * don't hold those caps, then we need to check whether
 630                          * the file is either opened or mmaped
 631                          */
 632                         if ((issued & (CEPH_CAP_FILE_CACHE|
 633                                        CEPH_CAP_FILE_BUFFER)) ||
 634                             mapping_mapped(inode->i_mapping) ||
 635                             __ceph_caps_file_wanted(ci)) {
 636                                 ci->i_truncate_pending++;
 637                                 queue_trunc = 1;
 638                         }
 639                 }
 640         }
 641         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 642             ci->i_truncate_size != truncate_size) {
 643                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 644                      truncate_size);
 645                 ci->i_truncate_size = truncate_size;
 646         }
 647
 648         if (queue_trunc)
 649                 ceph_fscache_invalidate(inode);
 650
 651         return queue_trunc;
 652 }
 653
 654 void ceph_fill_file_time(struct inode *inode, int issued,
 655                          u64 time_warp_seq, struct timespec64 *ctime,
 656                          struct timespec64 *mtime, struct timespec64 *atime)
 657 {
 658         struct ceph_inode_info *ci = ceph_inode(inode);
 659         int warn = 0;
 660
 661         if (issued & (CEPH_CAP_FILE_EXCL|
 662                       CEPH_CAP_FILE_WR|
 663                       CEPH_CAP_FILE_BUFFER|
 664                       CEPH_CAP_AUTH_EXCL|
 665                       CEPH_CAP_XATTR_EXCL)) {
 666                 if (ci->i_version == 0 ||
 667                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 668                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 669                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 670                              ctime->tv_sec, ctime->tv_nsec);
 671                         inode->i_ctime = *ctime;
 672                 }
 673                 if (ci->i_version == 0 ||
 674                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 675                         /* the MDS did a utimes() */
 676                         dout("mtime %lld.%09ld -> %lld.%09ld "
 677                              "tw %d -> %d\n",
 678                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 679                              mtime->tv_sec, mtime->tv_nsec,
 680                              ci->i_time_warp_seq, (int)time_warp_seq);
 681
 682                         inode->i_mtime = *mtime;
 683                         inode->i_atime = *atime;
 684                         ci->i_time_warp_seq = time_warp_seq;
 685                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 686                         /* nobody did utimes(); take the max */
 687                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 688                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 689                                      inode->i_mtime.tv_sec,
 690                                      inode->i_mtime.tv_nsec,
 691                                      mtime->tv_sec, mtime->tv_nsec);
 692                                 inode->i_mtime = *mtime;
 693                         }
 694                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 695                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 696                                      inode->i_atime.tv_sec,
 697                                      inode->i_atime.tv_nsec,
 698                                      atime->tv_sec, atime->tv_nsec);
 699                                 inode->i_atime = *atime;
 700                         }
 701                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 702                         /* we did a utimes(); ignore mds values */
 703                 } else {
 704                         warn = 1;
 705                 }
 706         } else {
 707                 /* we have no write|excl caps; whatever the MDS says is true */
 708                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 709                         inode->i_ctime = *ctime;
 710                         inode->i_mtime = *mtime;
 711                         inode->i_atime = *atime;
 712                         ci->i_time_warp_seq = time_warp_seq;
 713                 } else {
 714                         warn = 1;
 715                 }
 716         }
 717         if (warn) /* time_warp_seq shouldn't go backwards */
 718                 dout("%p mds time_warp_seq %llu < %u\n",
 719                      inode, time_warp_seq, ci->i_time_warp_seq);
 720 }
 721
 722 /*
 723  * Populate an inode based on info from mds.  May be called on new or
 724  * existing inodes.
 725  */
 726 static int fill_inode(struct inode *inode, struct page *locked_page,
 727                       struct ceph_mds_reply_info_in *iinfo,
 728                       struct ceph_mds_reply_dirfrag *dirinfo,
 729                       struct ceph_mds_session *session,
 730                       unsigned long ttl_from, int cap_fmode,
 731                       struct ceph_cap_reservation *caps_reservation)
 732 {
 733         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 734         struct ceph_mds_reply_inode *info = iinfo->in;
 735         struct ceph_inode_info *ci = ceph_inode(inode);
 736         int issued, new_issued, info_caps;
 737         struct timespec64 mtime, atime, ctime;
 738         struct ceph_buffer *xattr_blob = NULL;
 739         struct ceph_string *pool_ns = NULL;
 740         struct ceph_cap *new_cap = NULL;
 741         int err = 0;
 742         bool wake = false;
 743         bool queue_trunc = false;
 744         bool new_version = false;
 745         bool fill_inline = false;
 746
 747         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 748              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 749              ci->i_version);
 750
 751         info_caps = le32_to_cpu(info->cap.caps);
 752
 753         /* prealloc new cap struct */
 754         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
 755                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 756
 757         /*
 758          * prealloc xattr data, if it looks like we'll need it.  only
 759          * if len > 4 (meaning there are actually xattrs; the first 4
 760          * bytes are the xattr count).
 761          */
 762         if (iinfo->xattr_len > 4) {
 763                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 764                 if (!xattr_blob)
 765                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 766                                iinfo->xattr_len);
 767         }
 768
 769         if (iinfo->pool_ns_len > 0)
 770                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 771                                                      iinfo->pool_ns_len);
 772
 773         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 774                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 775
 776         spin_lock(&ci->i_ceph_lock);
 777
 778         /*
 779          * provided version will be odd if inode value is projected,
 780          * even if stable.  skip the update if we have newer stable
 781          * info (ours>=theirs, e.g. due to racing mds replies), unless
 782          * we are getting projected (unstable) info (in which case the
 783          * version is odd, and we want ours>theirs).
 784          *   us   them
 785          *   2    2     skip
 786          *   3    2     skip
 787          *   3    3     update
 788          */
 789         if (ci->i_version == 0 ||
 790             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 791              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 792                 new_version = true;
 793
 794         /* Update change_attribute */
 795         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 796
 797         __ceph_caps_issued(ci, &issued);
 798         issued |= __ceph_caps_dirty(ci);
 799         new_issued = ~issued & info_caps;
 800
 801         /* update inode */
 802         inode->i_rdev = le32_to_cpu(info->rdev);
 803         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 804
 805         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 806
 807         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 808             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 809                 inode->i_mode = le32_to_cpu(info->mode);
 810                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 811                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 812                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 813                      from_kuid(&init_user_ns, inode->i_uid),
 814                      from_kgid(&init_user_ns, inode->i_gid));
 815                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 816                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 817         }
 818
 819         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 820             (issued & CEPH_CAP_LINK_EXCL) == 0)
 821                 set_nlink(inode, le32_to_cpu(info->nlink));
 822
 823         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 824                 /* be careful with mtime, atime, size */
 825                 ceph_decode_timespec64(&atime, &info->atime);
 826                 ceph_decode_timespec64(&mtime, &info->mtime);
 827                 ceph_decode_timespec64(&ctime, &info->ctime);
 828                 ceph_fill_file_time(inode, issued,
 829                                 le32_to_cpu(info->time_warp_seq),
 830                                 &ctime, &mtime, &atime);
 831         }
 832
 833         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 834                 ci->i_files = le64_to_cpu(info->files);
 835                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 836         }
 837
 838         if (new_version ||
 839             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 840                 s64 old_pool = ci->i_layout.pool_id;
 841                 struct ceph_string *old_ns;
 842
 843                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 844                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 845                                         lockdep_is_held(&ci->i_ceph_lock));
 846                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 847
 848                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 849                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 850
 851                 pool_ns = old_ns;
 852
 853                 queue_trunc = ceph_fill_file_size(inode, issued,
 854                                         le32_to_cpu(info->truncate_seq),
 855                                         le64_to_cpu(info->truncate_size),
 856                                         le64_to_cpu(info->size));
 857                 /* only update max_size on auth cap */
 858                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 859                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 860                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 861                                         le64_to_cpu(info->max_size));
 862                         ci->i_max_size = le64_to_cpu(info->max_size);
 863                 }
 864         }
 865
 866         /* layout and rstat are not tracked by capability, update them if
 867          * the inode info is from auth mds */
 868         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 869                 if (S_ISDIR(inode->i_mode)) {
 870                         ci->i_dir_layout = iinfo->dir_layout;
 871                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 872                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 873                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 874                         ci->i_dir_pin = iinfo->dir_pin;
 875                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 876                 }
 877         }
 878
 879         /* xattrs */
 880         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 881         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 882             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 883                 if (ci->i_xattrs.blob)
 884                         ceph_buffer_put(ci->i_xattrs.blob);
 885                 ci->i_xattrs.blob = xattr_blob;
 886                 if (xattr_blob)
 887                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 888                                iinfo->xattr_data, iinfo->xattr_len);
 889                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 890                 ceph_forget_all_cached_acls(inode);
 891                 ceph_security_invalidate_secctx(inode);
 892                 xattr_blob = NULL;
 893         }
 894
 895         /* finally update i_version */
 896         if (le64_to_cpu(info->version) > ci->i_version)
 897                 ci->i_version = le64_to_cpu(info->version);
 898
 899         inode->i_mapping->a_ops = &ceph_aops;
 900
 901         switch (inode->i_mode & S_IFMT) {
 902         case S_IFIFO:
 903         case S_IFBLK:
 904         case S_IFCHR:
 905         case S_IFSOCK:
 906                 inode->i_blkbits = PAGE_SHIFT;
 907                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 908                 inode->i_op = &ceph_file_iops;
 909                 break;
 910         case S_IFREG:
 911                 inode->i_op = &ceph_file_iops;
 912                 inode->i_fop = &ceph_file_fops;
 913                 break;
 914         case S_IFLNK:
 915                 inode->i_op = &ceph_symlink_iops;
 916                 if (!ci->i_symlink) {
 917                         u32 symlen = iinfo->symlink_len;
 918                         char *sym;
 919
 920                         spin_unlock(&ci->i_ceph_lock);
 921
 922                         if (symlen != i_size_read(inode)) {
 923                                 pr_err("fill_inode %llx.%llx BAD symlink "
 924                                         "size %lld\n", ceph_vinop(inode),
 925                                         i_size_read(inode));
 926                                 i_size_write(inode, symlen);
 927                                 inode->i_blocks = calc_inode_blocks(symlen);
 928                         }
 929
 930                         err = -ENOMEM;
 931                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 932                         if (!sym)
 933                                 goto out;
 934
 935                         spin_lock(&ci->i_ceph_lock);
 936                         if (!ci->i_symlink)
 937                                 ci->i_symlink = sym;
 938                         else
 939                                 kfree(sym); /* lost a race */
 940                 }
 941                 inode->i_link = ci->i_symlink;
 942                 break;
 943         case S_IFDIR:
 944                 inode->i_op = &ceph_dir_iops;
 945                 inode->i_fop = &ceph_dir_fops;
 946                 break;
 947         default:
 948                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 949                        ceph_vinop(inode), inode->i_mode);
 950         }
 951
 952         /* were we issued a capability? */
 953         if (info_caps) {
 954                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 955                         ceph_add_cap(inode, session,
 956                                      le64_to_cpu(info->cap.cap_id),
 957                                      cap_fmode, info_caps,
 958                                      le32_to_cpu(info->cap.wanted),
 959                                      le32_to_cpu(info->cap.seq),
 960                                      le32_to_cpu(info->cap.mseq),
 961                                      le64_to_cpu(info->cap.realm),
 962                                      info->cap.flags, &new_cap);
 963
 964                         /* set dir completion flag? */
 965                         if (S_ISDIR(inode->i_mode) &&
 966                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 967                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 968                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 969                             !__ceph_dir_is_complete(ci)) {
 970                                 dout(" marking %p complete (empty)\n", inode);
 971                                 i_size_write(inode, 0);
 972                                 __ceph_dir_set_complete(ci,
 973                                         atomic64_read(&ci->i_release_count),
 974                                         atomic64_read(&ci->i_ordered_count));
 975                         }
 976
 977                         wake = true;
 978                 } else {
 979                         dout(" %p got snap_caps %s\n", inode,
 980                              ceph_cap_string(info_caps));
 981                         ci->i_snap_caps |= info_caps;
 982                         if (cap_fmode >= 0)
 983                                 __ceph_get_fmode(ci, cap_fmode);
 984                 }
 985         } else if (cap_fmode >= 0) {
 986                 pr_warn("mds issued no caps on %llx.%llx\n",
 987                            ceph_vinop(inode));
 988                 __ceph_get_fmode(ci, cap_fmode);
 989         }
 990
 991         if (iinfo->inline_version > 0 &&
 992             iinfo->inline_version >= ci->i_inline_version) {
 993                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 994                 ci->i_inline_version = iinfo->inline_version;
 995                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 996                     (locked_page || (info_caps & cache_caps)))
 997                         fill_inline = true;
 998         }
 999
1000         spin_unlock(&ci->i_ceph_lock);
1001
1002         if (fill_inline)
1003                 ceph_fill_inline_data(inode, locked_page,
1004                                       iinfo->inline_data, iinfo->inline_len);
1005
1006         if (wake)
1007                 wake_up_all(&ci->i_cap_wq);
1008
1009         /* queue truncate if we saw i_size decrease */
1010         if (queue_trunc)
1011                 ceph_queue_vmtruncate(inode);
1012
1013         /* populate frag tree */
1014         if (S_ISDIR(inode->i_mode))
1015                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1016
1017         /* update delegation info? */
1018         if (dirinfo)
1019                 ceph_fill_dirfrag(inode, dirinfo);
1020
1021         err = 0;
1022 out:
1023         if (new_cap)
1024                 ceph_put_cap(mdsc, new_cap);
1025         if (xattr_blob)
1026                 ceph_buffer_put(xattr_blob);
1027         ceph_put_string(pool_ns);
1028         return err;
1029 }
1030
1031 /*
1032  * caller should hold session s_mutex and dentry->d_lock.
1033  */
1034 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1035                                   struct ceph_mds_reply_lease *lease,
1036                                   struct ceph_mds_session *session,
1037                                   unsigned long from_time,
1038                                   struct ceph_mds_session **old_lease_session)
1039 {
1040         struct ceph_dentry_info *di = ceph_dentry(dentry);
1041         long unsigned duration = le32_to_cpu(lease->duration_ms);
1042         long unsigned ttl = from_time + (duration * HZ) / 1000;
1043         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1044
1045         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1046              dentry, duration, ttl);
1047
1048         /* only track leases on regular dentries */
1049         if (ceph_snap(dir) != CEPH_NOSNAP)
1050                 return;
1051
1052         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1053         if (duration == 0) {
1054                 __ceph_dentry_dir_lease_touch(di);
1055                 return;
1056         }
1057
1058         if (di->lease_gen == session->s_cap_gen &&
1059             time_before(ttl, di->time))
1060                 return;  /* we already have a newer lease. */
1061
1062         if (di->lease_session && di->lease_session != session) {
1063                 *old_lease_session = di->lease_session;
1064                 di->lease_session = NULL;
1065         }
1066
1067         if (!di->lease_session)
1068                 di->lease_session = ceph_get_mds_session(session);
1069         di->lease_gen = session->s_cap_gen;
1070         di->lease_seq = le32_to_cpu(lease->seq);
1071         di->lease_renew_after = half_ttl;
1072         di->lease_renew_from = 0;
1073         di->time = ttl;
1074
1075         __ceph_dentry_lease_touch(di);
1076 }
1077
1078 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1079                                         struct ceph_mds_reply_lease *lease,
1080                                         struct ceph_mds_session *session,
1081                                         unsigned long from_time)
1082 {
1083         struct ceph_mds_session *old_lease_session = NULL;
1084         spin_lock(&dentry->d_lock);
1085         __update_dentry_lease(dir, dentry, lease, session, from_time,
1086                               &old_lease_session);
1087         spin_unlock(&dentry->d_lock);
1088         if (old_lease_session)
1089                 ceph_put_mds_session(old_lease_session);
1090 }
1091
1092 /*
1093  * update dentry lease without having parent inode locked
1094  */
1095 static void update_dentry_lease_careful(struct dentry *dentry,
1096                                         struct ceph_mds_reply_lease *lease,
1097                                         struct ceph_mds_session *session,
1098                                         unsigned long from_time,
1099                                         char *dname, u32 dname_len,
1100                                         struct ceph_vino *pdvino,
1101                                         struct ceph_vino *ptvino)
1102
1103 {
1104         struct inode *dir;
1105         struct ceph_mds_session *old_lease_session = NULL;
1106
1107         spin_lock(&dentry->d_lock);
1108         /* make sure dentry's name matches target */
1109         if (dentry->d_name.len != dname_len ||
1110             memcmp(dentry->d_name.name, dname, dname_len))
1111                 goto out_unlock;
1112
1113         dir = d_inode(dentry->d_parent);
1114         /* make sure parent matches dvino */
1115         if (!ceph_ino_compare(dir, pdvino))
1116                 goto out_unlock;
1117
1118         /* make sure dentry's inode matches target. NULL ptvino means that
1119          * we expect a negative dentry */
1120         if (ptvino) {
1121                 if (d_really_is_negative(dentry))
1122                         goto out_unlock;
1123                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1124                         goto out_unlock;
1125         } else {
1126                 if (d_really_is_positive(dentry))
1127                         goto out_unlock;
1128         }
1129
1130         __update_dentry_lease(dir, dentry, lease, session,
1131                               from_time, &old_lease_session);
1132 out_unlock:
1133         spin_unlock(&dentry->d_lock);
1134         if (old_lease_session)
1135                 ceph_put_mds_session(old_lease_session);
1136 }
1137
1138 /*
1139  * splice a dentry to an inode.
1140  * caller must hold directory i_mutex for this to be safe.
1141  */
1142 static int splice_dentry(struct dentry **pdn, struct inode *in)
1143 {
1144         struct dentry *dn = *pdn;
1145         struct dentry *realdn;
1146
1147         BUG_ON(d_inode(dn));
1148
1149         if (S_ISDIR(in->i_mode)) {
1150                 /* If inode is directory, d_splice_alias() below will remove
1151                  * 'realdn' from its origin parent. We need to ensure that
1152                  * origin parent's readdir cache will not reference 'realdn'
1153                  */
1154                 realdn = d_find_any_alias(in);
1155                 if (realdn) {
1156                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1157                         spin_lock(&realdn->d_lock);
1158
1159                         realdn->d_op->d_prune(realdn);
1160
1161                         di->time = jiffies;
1162                         di->lease_shared_gen = 0;
1163                         di->offset = 0;
1164
1165                         spin_unlock(&realdn->d_lock);
1166                         dput(realdn);
1167                 }
1168         }
1169
1170         /* dn must be unhashed */
1171         if (!d_unhashed(dn))
1172                 d_drop(dn);
1173         realdn = d_splice_alias(in, dn);
1174         if (IS_ERR(realdn)) {
1175                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1176                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1177                 return PTR_ERR(realdn);
1178         }
1179
1180         if (realdn) {
1181                 dout("dn %p (%d) spliced with %p (%d) "
1182                      "inode %p ino %llx.%llx\n",
1183                      dn, d_count(dn),
1184                      realdn, d_count(realdn),
1185                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1186                 dput(dn);
1187                 *pdn = realdn;
1188         } else {
1189                 BUG_ON(!ceph_dentry(dn));
1190                 dout("dn %p attached to %p ino %llx.%llx\n",
1191                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1192         }
1193         return 0;
1194 }
1195
1196 /*
1197  * Incorporate results into the local cache.  This is either just
1198  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1199  * after a lookup).
1200  *
1201  * A reply may contain
1202  *         a directory inode along with a dentry.
1203  *  and/or a target inode
1204  *
1205  * Called with snap_rwsem (read).
1206  */
1207 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1208 {
1209         struct ceph_mds_session *session = req->r_session;
1210         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1211         struct inode *in = NULL;
1212         struct ceph_vino tvino, dvino;
1213         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1214         int err = 0;
1215
1216         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1217              rinfo->head->is_dentry, rinfo->head->is_target);
1218
1219         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1220                 dout("fill_trace reply is empty!\n");
1221                 if (rinfo->head->result == 0 && req->r_parent)
1222                         ceph_invalidate_dir_request(req);
1223                 return 0;
1224         }
1225
1226         if (rinfo->head->is_dentry) {
1227                 struct inode *dir = req->r_parent;
1228
1229                 if (dir) {
1230                         err = fill_inode(dir, NULL,
1231                                          &rinfo->diri, rinfo->dirfrag,
1232                                          session, req->r_request_started, -1,
1233                                          &req->r_caps_reservation);
1234                         if (err < 0)
1235                                 goto done;
1236                 } else {
1237                         WARN_ON_ONCE(1);
1238                 }
1239
1240                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1241                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1242                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1243                         struct qstr dname;
1244                         struct dentry *dn, *parent;
1245
1246                         BUG_ON(!rinfo->head->is_target);
1247                         BUG_ON(req->r_dentry);
1248
1249                         parent = d_find_any_alias(dir);
1250                         BUG_ON(!parent);
1251
1252                         dname.name = rinfo->dname;
1253                         dname.len = rinfo->dname_len;
1254                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1255                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1256                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1257 retry_lookup:
1258                         dn = d_lookup(parent, &dname);
1259                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1260                              parent, dname.len, dname.name, dn);
1261
1262                         if (!dn) {
1263                                 dn = d_alloc(parent, &dname);
1264                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1265                                      dname.len, dname.name, dn);
1266                                 if (!dn) {
1267                                         dput(parent);
1268                                         err = -ENOMEM;
1269                                         goto done;
1270                                 }
1271                                 err = 0;
1272                         } else if (d_really_is_positive(dn) &&
1273                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1274                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1275                                 dout(" dn %p points to wrong inode %p\n",
1276                                      dn, d_inode(dn));
1277                                 ceph_dir_clear_ordered(dir);
1278                                 d_delete(dn);
1279                                 dput(dn);
1280                                 goto retry_lookup;
1281                         }
1282
1283                         req->r_dentry = dn;
1284                         dput(parent);
1285                 }
1286         }
1287
1288         if (rinfo->head->is_target) {
1289                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1290                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1291
1292                 in = ceph_get_inode(sb, tvino);
1293                 if (IS_ERR(in)) {
1294                         err = PTR_ERR(in);
1295                         goto done;
1296                 }
1297                 req->r_target_inode = in;
1298
1299                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1300                                 session, req->r_request_started,
1301                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1302                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1303                                 &req->r_caps_reservation);
1304                 if (err < 0) {
1305                         pr_err("fill_inode badness %p %llx.%llx\n",
1306                                 in, ceph_vinop(in));
1307                         goto done;
1308                 }
1309         }
1310
1311         /*
1312          * ignore null lease/binding on snapdir ENOENT, or else we
1313          * will have trouble splicing in the virtual snapdir later
1314          */
1315         if (rinfo->head->is_dentry &&
1316             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1317             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1318             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1319                                                fsc->mount_options->snapdir_name,
1320                                                req->r_dentry->d_name.len))) {
1321                 /*
1322                  * lookup link rename   : null -> possibly existing inode
1323                  * mknod symlink mkdir  : null -> new inode
1324                  * unlink               : linked -> null
1325                  */
1326                 struct inode *dir = req->r_parent;
1327                 struct dentry *dn = req->r_dentry;
1328                 bool have_dir_cap, have_lease;
1329
1330                 BUG_ON(!dn);
1331                 BUG_ON(!dir);
1332                 BUG_ON(d_inode(dn->d_parent) != dir);
1333
1334                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1335                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1336
1337                 BUG_ON(ceph_ino(dir) != dvino.ino);
1338                 BUG_ON(ceph_snap(dir) != dvino.snap);
1339
1340                 /* do we have a lease on the whole dir? */
1341                 have_dir_cap =
1342                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1343                          CEPH_CAP_FILE_SHARED);
1344
1345                 /* do we have a dn lease? */
1346                 have_lease = have_dir_cap ||
1347                         le32_to_cpu(rinfo->dlease->duration_ms);
1348                 if (!have_lease)
1349                         dout("fill_trace  no dentry lease or dir cap\n");
1350
1351                 /* rename? */
1352                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1353                         struct inode *olddir = req->r_old_dentry_dir;
1354                         BUG_ON(!olddir);
1355
1356                         dout(" src %p '%pd' dst %p '%pd'\n",
1357                              req->r_old_dentry,
1358                              req->r_old_dentry,
1359                              dn, dn);
1360                         dout("fill_trace doing d_move %p -> %p\n",
1361                              req->r_old_dentry, dn);
1362
1363                         /* d_move screws up sibling dentries' offsets */
1364                         ceph_dir_clear_ordered(dir);
1365                         ceph_dir_clear_ordered(olddir);
1366
1367                         d_move(req->r_old_dentry, dn);
1368                         dout(" src %p '%pd' dst %p '%pd'\n",
1369                              req->r_old_dentry,
1370                              req->r_old_dentry,
1371                              dn, dn);
1372
1373                         /* ensure target dentry is invalidated, despite
1374                            rehashing bug in vfs_rename_dir */
1375                         ceph_invalidate_dentry_lease(dn);
1376
1377                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1378                              ceph_dentry(req->r_old_dentry)->offset);
1379
1380                         /* swap r_dentry and r_old_dentry in case that
1381                          * splice_dentry() gets called later. This is safe
1382                          * because no other place will use them */
1383                         req->r_dentry = req->r_old_dentry;
1384                         req->r_old_dentry = dn;
1385                         dn = req->r_dentry;
1386                 }
1387
1388                 /* null dentry? */
1389                 if (!rinfo->head->is_target) {
1390                         dout("fill_trace null dentry\n");
1391                         if (d_really_is_positive(dn)) {
1392                                 dout("d_delete %p\n", dn);
1393                                 ceph_dir_clear_ordered(dir);
1394                                 d_delete(dn);
1395                         } else if (have_lease) {
1396                                 if (d_unhashed(dn))
1397                                         d_add(dn, NULL);
1398                                 update_dentry_lease(dir, dn,
1399                                                     rinfo->dlease, session,
1400                                                     req->r_request_started);
1401                         }
1402                         goto done;
1403                 }
1404
1405                 /* attach proper inode */
1406                 if (d_really_is_negative(dn)) {
1407                         ceph_dir_clear_ordered(dir);
1408                         ihold(in);
1409                         err = splice_dentry(&req->r_dentry, in);
1410                         if (err < 0)
1411                                 goto done;
1412                         dn = req->r_dentry;  /* may have spliced */
1413                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1414                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1415                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1416                              ceph_vinop(in));
1417                         d_invalidate(dn);
1418                         have_lease = false;
1419                 }
1420
1421                 if (have_lease) {
1422                         update_dentry_lease(dir, dn,
1423                                             rinfo->dlease, session,
1424                                             req->r_request_started);
1425                 }
1426                 dout(" final dn %p\n", dn);
1427         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1428                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1429                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1430                 struct inode *dir = req->r_parent;
1431
1432                 /* fill out a snapdir LOOKUPSNAP dentry */
1433                 BUG_ON(!dir);
1434                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1435                 BUG_ON(!req->r_dentry);
1436                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1437                 ceph_dir_clear_ordered(dir);
1438                 ihold(in);
1439                 err = splice_dentry(&req->r_dentry, in);
1440                 if (err < 0)
1441                         goto done;
1442         } else if (rinfo->head->is_dentry && req->r_dentry) {
1443                 /* parent inode is not locked, be carefull */
1444                 struct ceph_vino *ptvino = NULL;
1445                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1446                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1447                 if (rinfo->head->is_target) {
1448                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1449                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1450                         ptvino = &tvino;
1451                 }
1452                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1453                                             session, req->r_request_started,
1454                                             rinfo->dname, rinfo->dname_len,
1455                                             &dvino, ptvino);
1456         }
1457 done:
1458         dout("fill_trace done err=%d\n", err);
1459         return err;
1460 }
1461
1462 /*
1463  * Prepopulate our cache with readdir results, leases, etc.
1464  */
1465 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1466                                            struct ceph_mds_session *session)
1467 {
1468         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1469         int i, err = 0;
1470
1471         for (i = 0; i < rinfo->dir_nr; i++) {
1472                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1473                 struct ceph_vino vino;
1474                 struct inode *in;
1475                 int rc;
1476
1477                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1478                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1479
1480                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1481                 if (IS_ERR(in)) {
1482                         err = PTR_ERR(in);
1483                         dout("new_inode badness got %d\n", err);
1484                         continue;
1485                 }
1486                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1487                                 req->r_request_started, -1,
1488                                 &req->r_caps_reservation);
1489                 if (rc < 0) {
1490                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1491                         err = rc;
1492                 }
1493                 /* avoid calling iput_final() in mds dispatch threads */
1494                 ceph_async_iput(in);
1495         }
1496
1497         return err;
1498 }
1499
1500 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1501 {
1502         if (ctl->page) {
1503                 kunmap(ctl->page);
1504                 put_page(ctl->page);
1505                 ctl->page = NULL;
1506         }
1507 }
1508
1509 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1510                               struct ceph_readdir_cache_control *ctl,
1511                               struct ceph_mds_request *req)
1512 {
1513         struct ceph_inode_info *ci = ceph_inode(dir);
1514         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1515         unsigned idx = ctl->index % nsize;
1516         pgoff_t pgoff = ctl->index / nsize;
1517
1518         if (!ctl->page || pgoff != page_index(ctl->page)) {
1519                 ceph_readdir_cache_release(ctl);
1520                 if (idx == 0)
1521                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1522                 else
1523                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1524                 if (!ctl->page) {
1525                         ctl->index = -1;
1526                         return idx == 0 ? -ENOMEM : 0;
1527                 }
1528                 /* reading/filling the cache are serialized by
1529                  * i_mutex, no need to use page lock */
1530                 unlock_page(ctl->page);
1531                 ctl->dentries = kmap(ctl->page);
1532                 if (idx == 0)
1533                         memset(ctl->dentries, 0, PAGE_SIZE);
1534         }
1535
1536         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1537             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1538                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1539                 ctl->dentries[idx] = dn;
1540                 ctl->index++;
1541         } else {
1542                 dout("disable readdir cache\n");
1543                 ctl->index = -1;
1544         }
1545         return 0;
1546 }
1547
1548 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1549                              struct ceph_mds_session *session)
1550 {
1551         struct dentry *parent = req->r_dentry;
1552         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1553         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1554         struct qstr dname;
1555         struct dentry *dn;
1556         struct inode *in;
1557         int err = 0, skipped = 0, ret, i;
1558         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1559         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1560         u32 last_hash = 0;
1561         u32 fpos_offset;
1562         struct ceph_readdir_cache_control cache_ctl = {};
1563
1564         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1565                 return readdir_prepopulate_inodes_only(req, session);
1566
1567         if (rinfo->hash_order) {
1568                 if (req->r_path2) {
1569                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1570                                                   req->r_path2,
1571                                                   strlen(req->r_path2));
1572                         last_hash = ceph_frag_value(last_hash);
1573                 } else if (rinfo->offset_hash) {
1574                         /* mds understands offset_hash */
1575                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1576                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1577                 }
1578         }
1579
1580         if (rinfo->dir_dir &&
1581             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1582                 dout("readdir_prepopulate got new frag %x -> %x\n",
1583                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1584                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1585                 if (!rinfo->hash_order)
1586                         req->r_readdir_offset = 2;
1587         }
1588
1589         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1590                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1591                      rinfo->dir_nr, parent);
1592         } else {
1593                 dout("readdir_prepopulate %d items under dn %p\n",
1594                      rinfo->dir_nr, parent);
1595                 if (rinfo->dir_dir)
1596                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1597
1598                 if (ceph_frag_is_leftmost(frag) &&
1599                     req->r_readdir_offset == 2 &&
1600                     !(rinfo->hash_order && last_hash)) {
1601                         /* note dir version at start of readdir so we can
1602                          * tell if any dentries get dropped */
1603                         req->r_dir_release_cnt =
1604                                 atomic64_read(&ci->i_release_count);
1605                         req->r_dir_ordered_cnt =
1606                                 atomic64_read(&ci->i_ordered_count);
1607                         req->r_readdir_cache_idx = 0;
1608                 }
1609         }
1610
1611         cache_ctl.index = req->r_readdir_cache_idx;
1612         fpos_offset = req->r_readdir_offset;
1613
1614         /* FIXME: release caps/leases if error occurs */
1615         for (i = 0; i < rinfo->dir_nr; i++) {
1616                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1617                 struct ceph_vino tvino;
1618
1619                 dname.name = rde->name;
1620                 dname.len = rde->name_len;
1621                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1622
1623                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1624                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1625
1626                 if (rinfo->hash_order) {
1627                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1628                                                  rde->name, rde->name_len);
1629                         hash = ceph_frag_value(hash);
1630                         if (hash != last_hash)
1631                                 fpos_offset = 2;
1632                         last_hash = hash;
1633                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1634                 } else {
1635                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1636                 }
1637
1638 retry_lookup:
1639                 dn = d_lookup(parent, &dname);
1640                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1641                      parent, dname.len, dname.name, dn);
1642
1643                 if (!dn) {
1644                         dn = d_alloc(parent, &dname);
1645                         dout("d_alloc %p '%.*s' = %p\n", parent,
1646                              dname.len, dname.name, dn);
1647                         if (!dn) {
1648                                 dout("d_alloc badness\n");
1649                                 err = -ENOMEM;
1650                                 goto out;
1651                         }
1652                 } else if (d_really_is_positive(dn) &&
1653                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1654                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1655                         struct ceph_dentry_info *di = ceph_dentry(dn);
1656                         dout(" dn %p points to wrong inode %p\n",
1657                              dn, d_inode(dn));
1658
1659                         spin_lock(&dn->d_lock);
1660                         if (di->offset > 0 &&
1661                             di->lease_shared_gen ==
1662                             atomic_read(&ci->i_shared_gen)) {
1663                                 __ceph_dir_clear_ordered(ci);
1664                                 di->offset = 0;
1665                         }
1666                         spin_unlock(&dn->d_lock);
1667
1668                         d_delete(dn);
1669                         dput(dn);
1670                         goto retry_lookup;
1671                 }
1672
1673                 /* inode */
1674                 if (d_really_is_positive(dn)) {
1675                         in = d_inode(dn);
1676                 } else {
1677                         in = ceph_get_inode(parent->d_sb, tvino);
1678                         if (IS_ERR(in)) {
1679                                 dout("new_inode badness\n");
1680                                 d_drop(dn);
1681                                 dput(dn);
1682                                 err = PTR_ERR(in);
1683                                 goto out;
1684                         }
1685                 }
1686
1687                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1688                                  req->r_request_started, -1,
1689                                  &req->r_caps_reservation);
1690                 if (ret < 0) {
1691                         pr_err("fill_inode badness on %p\n", in);
1692                         if (d_really_is_negative(dn)) {
1693                                 /* avoid calling iput_final() in mds
1694                                  * dispatch threads */
1695                                 ceph_async_iput(in);
1696                         }
1697                         d_drop(dn);
1698                         err = ret;
1699                         goto next_item;
1700                 }
1701
1702                 if (d_really_is_negative(dn)) {
1703                         if (ceph_security_xattr_deadlock(in)) {
1704                                 dout(" skip splicing dn %p to inode %p"
1705                                      " (security xattr deadlock)\n", dn, in);
1706                                 ceph_async_iput(in);
1707                                 skipped++;
1708                                 goto next_item;
1709                         }
1710
1711                         err = splice_dentry(&dn, in);
1712                         if (err < 0)
1713                                 goto next_item;
1714                 }
1715
1716                 ceph_dentry(dn)->offset = rde->offset;
1717
1718                 update_dentry_lease(d_inode(parent), dn,
1719                                     rde->lease, req->r_session,
1720                                     req->r_request_started);
1721
1722                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1723                         ret = fill_readdir_cache(d_inode(parent), dn,
1724                                                  &cache_ctl, req);
1725                         if (ret < 0)
1726                                 err = ret;
1727                 }
1728 next_item:
1729                 dput(dn);
1730         }
1731 out:
1732         if (err == 0 && skipped == 0) {
1733                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1734                 req->r_readdir_cache_idx = cache_ctl.index;
1735         }
1736         ceph_readdir_cache_release(&cache_ctl);
1737         dout("readdir_prepopulate done\n");
1738         return err;
1739 }
1740
1741 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1742 {
1743         struct ceph_inode_info *ci = ceph_inode(inode);
1744         bool ret;
1745
1746         spin_lock(&ci->i_ceph_lock);
1747         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1748         i_size_write(inode, size);
1749         inode->i_blocks = calc_inode_blocks(size);
1750
1751         ret = __ceph_should_report_size(ci);
1752
1753         spin_unlock(&ci->i_ceph_lock);
1754         return ret;
1755 }
1756
1757 /*
1758  * Put reference to inode, but avoid calling iput_final() in current thread.
1759  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1760  * some contexts.
1761  */
1762 void ceph_async_iput(struct inode *inode)
1763 {
1764         if (!inode)
1765                 return;
1766         for (;;) {
1767                 if (atomic_add_unless(&inode->i_count, -1, 1))
1768                         break;
1769                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1770                                &ceph_inode(inode)->i_work))
1771                         break;
1772                 /* queue work failed, i_count must be at least 2 */
1773         }
1774 }
1775
1776 /*
1777  * Write back inode data in a worker thread.  (This can't be done
1778  * in the message handler context.)
1779  */
1780 void ceph_queue_writeback(struct inode *inode)
1781 {
1782         struct ceph_inode_info *ci = ceph_inode(inode);
1783         set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1784
1785         ihold(inode);
1786         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1787                        &ci->i_work)) {
1788                 dout("ceph_queue_writeback %p\n", inode);
1789         } else {
1790                 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1791                      inode, ci->i_work_mask);
1792                 iput(inode);
1793         }
1794 }
1795
1796 /*
1797  * queue an async invalidation
1798  */
1799 void ceph_queue_invalidate(struct inode *inode)
1800 {
1801         struct ceph_inode_info *ci = ceph_inode(inode);
1802         set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1803
1804         ihold(inode);
1805         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1806                        &ceph_inode(inode)->i_work)) {
1807                 dout("ceph_queue_invalidate %p\n", inode);
1808         } else {
1809                 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1810                      inode, ci->i_work_mask);
1811                 iput(inode);
1812         }
1813 }
1814
1815 /*
1816  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1817  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1818  */
1819 void ceph_queue_vmtruncate(struct inode *inode)
1820 {
1821         struct ceph_inode_info *ci = ceph_inode(inode);
1822         set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1823
1824         ihold(inode);
1825         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1826                        &ci->i_work)) {
1827                 dout("ceph_queue_vmtruncate %p\n", inode);
1828         } else {
1829                 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1830                      inode, ci->i_work_mask);
1831                 iput(inode);
1832         }
1833 }
1834
1835 static void ceph_do_invalidate_pages(struct inode *inode)
1836 {
1837         struct ceph_inode_info *ci = ceph_inode(inode);
1838         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1839         u32 orig_gen;
1840         int check = 0;
1841
1842         mutex_lock(&ci->i_truncate_mutex);
1843
1844         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1845                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1846                                     inode, ceph_ino(inode));
1847                 mapping_set_error(inode->i_mapping, -EIO);
1848                 truncate_pagecache(inode, 0);
1849                 mutex_unlock(&ci->i_truncate_mutex);
1850                 goto out;
1851         }
1852
1853         spin_lock(&ci->i_ceph_lock);
1854         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1855              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1856         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1857                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1858                         check = 1;
1859                 spin_unlock(&ci->i_ceph_lock);
1860                 mutex_unlock(&ci->i_truncate_mutex);
1861                 goto out;
1862         }
1863         orig_gen = ci->i_rdcache_gen;
1864         spin_unlock(&ci->i_ceph_lock);
1865
1866         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1867                 pr_err("invalidate_pages %p fails\n", inode);
1868         }
1869
1870         spin_lock(&ci->i_ceph_lock);
1871         if (orig_gen == ci->i_rdcache_gen &&
1872             orig_gen == ci->i_rdcache_revoking) {
1873                 dout("invalidate_pages %p gen %d successful\n", inode,
1874                      ci->i_rdcache_gen);
1875                 ci->i_rdcache_revoking--;
1876                 check = 1;
1877         } else {
1878                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1879                      inode, orig_gen, ci->i_rdcache_gen,
1880                      ci->i_rdcache_revoking);
1881                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1882                         check = 1;
1883         }
1884         spin_unlock(&ci->i_ceph_lock);
1885         mutex_unlock(&ci->i_truncate_mutex);
1886 out:
1887         if (check)
1888                 ceph_check_caps(ci, 0, NULL);
1889 }
1890
1891 /*
1892  * Make sure any pending truncation is applied before doing anything
1893  * that may depend on it.
1894  */
1895 void __ceph_do_pending_vmtruncate(struct inode *inode)
1896 {
1897         struct ceph_inode_info *ci = ceph_inode(inode);
1898         u64 to;
1899         int wrbuffer_refs, finish = 0;
1900
1901         mutex_lock(&ci->i_truncate_mutex);
1902 retry:
1903         spin_lock(&ci->i_ceph_lock);
1904         if (ci->i_truncate_pending == 0) {
1905                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1906                 spin_unlock(&ci->i_ceph_lock);
1907                 mutex_unlock(&ci->i_truncate_mutex);
1908                 return;
1909         }
1910
1911         /*
1912          * make sure any dirty snapped pages are flushed before we
1913          * possibly truncate them.. so write AND block!
1914          */
1915         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1916                 spin_unlock(&ci->i_ceph_lock);
1917                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1918                      inode);
1919                 filemap_write_and_wait_range(&inode->i_data, 0,
1920                                              inode->i_sb->s_maxbytes);
1921                 goto retry;
1922         }
1923
1924         /* there should be no reader or writer */
1925         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1926
1927         to = ci->i_truncate_size;
1928         wrbuffer_refs = ci->i_wrbuffer_ref;
1929         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1930              ci->i_truncate_pending, to);
1931         spin_unlock(&ci->i_ceph_lock);
1932
1933         truncate_pagecache(inode, to);
1934
1935         spin_lock(&ci->i_ceph_lock);
1936         if (to == ci->i_truncate_size) {
1937                 ci->i_truncate_pending = 0;
1938                 finish = 1;
1939         }
1940         spin_unlock(&ci->i_ceph_lock);
1941         if (!finish)
1942                 goto retry;
1943
1944         mutex_unlock(&ci->i_truncate_mutex);
1945
1946         if (wrbuffer_refs == 0)
1947                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1948
1949         wake_up_all(&ci->i_cap_wq);
1950 }
1951
1952 static void ceph_inode_work(struct work_struct *work)
1953 {
1954         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1955                                                  i_work);
1956         struct inode *inode = &ci->vfs_inode;
1957
1958         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1959                 dout("writeback %p\n", inode);
1960                 filemap_fdatawrite(&inode->i_data);
1961         }
1962         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1963                 ceph_do_invalidate_pages(inode);
1964
1965         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1966                 __ceph_do_pending_vmtruncate(inode);
1967
1968         iput(inode);
1969 }
1970
1971 /*
1972  * symlinks
1973  */
1974 static const struct inode_operations ceph_symlink_iops = {
1975         .get_link = simple_get_link,
1976         .setattr = ceph_setattr,
1977         .getattr = ceph_getattr,
1978         .listxattr = ceph_listxattr,
1979 };
1980
1981 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1982 {
1983         struct ceph_inode_info *ci = ceph_inode(inode);
1984         const unsigned int ia_valid = attr->ia_valid;
1985         struct ceph_mds_request *req;
1986         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1987         struct ceph_cap_flush *prealloc_cf;
1988         int issued;
1989         int release = 0, dirtied = 0;
1990         int mask = 0;
1991         int err = 0;
1992         int inode_dirty_flags = 0;
1993         bool lock_snap_rwsem = false;
1994
1995         prealloc_cf = ceph_alloc_cap_flush();
1996         if (!prealloc_cf)
1997                 return -ENOMEM;
1998
1999         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2000                                        USE_AUTH_MDS);
2001         if (IS_ERR(req)) {
2002                 ceph_free_cap_flush(prealloc_cf);
2003                 return PTR_ERR(req);
2004         }
2005
2006         spin_lock(&ci->i_ceph_lock);
2007         issued = __ceph_caps_issued(ci, NULL);
2008
2009         if (!ci->i_head_snapc &&
2010             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2011                 lock_snap_rwsem = true;
2012                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2013                         spin_unlock(&ci->i_ceph_lock);
2014                         down_read(&mdsc->snap_rwsem);
2015                         spin_lock(&ci->i_ceph_lock);
2016                         issued = __ceph_caps_issued(ci, NULL);
2017                 }
2018         }
2019
2020         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2021
2022         if (ia_valid & ATTR_UID) {
2023                 dout("setattr %p uid %d -> %d\n", inode,
2024                      from_kuid(&init_user_ns, inode->i_uid),
2025                      from_kuid(&init_user_ns, attr->ia_uid));
2026                 if (issued & CEPH_CAP_AUTH_EXCL) {
2027                         inode->i_uid = attr->ia_uid;
2028                         dirtied |= CEPH_CAP_AUTH_EXCL;
2029                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2030                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2031                         req->r_args.setattr.uid = cpu_to_le32(
2032                                 from_kuid(&init_user_ns, attr->ia_uid));
2033                         mask |= CEPH_SETATTR_UID;
2034                         release |= CEPH_CAP_AUTH_SHARED;
2035                 }
2036         }
2037         if (ia_valid & ATTR_GID) {
2038                 dout("setattr %p gid %d -> %d\n", inode,
2039                      from_kgid(&init_user_ns, inode->i_gid),
2040                      from_kgid(&init_user_ns, attr->ia_gid));
2041                 if (issued & CEPH_CAP_AUTH_EXCL) {
2042                         inode->i_gid = attr->ia_gid;
2043                         dirtied |= CEPH_CAP_AUTH_EXCL;
2044                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2045                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2046                         req->r_args.setattr.gid = cpu_to_le32(
2047                                 from_kgid(&init_user_ns, attr->ia_gid));
2048                         mask |= CEPH_SETATTR_GID;
2049                         release |= CEPH_CAP_AUTH_SHARED;
2050                 }
2051         }
2052         if (ia_valid & ATTR_MODE) {
2053                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2054                      attr->ia_mode);
2055                 if (issued & CEPH_CAP_AUTH_EXCL) {
2056                         inode->i_mode = attr->ia_mode;
2057                         dirtied |= CEPH_CAP_AUTH_EXCL;
2058                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2059                            attr->ia_mode != inode->i_mode) {
2060                         inode->i_mode = attr->ia_mode;
2061                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2062                         mask |= CEPH_SETATTR_MODE;
2063                         release |= CEPH_CAP_AUTH_SHARED;
2064                 }
2065         }
2066
2067         if (ia_valid & ATTR_ATIME) {
2068                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2069                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2070                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2071                 if (issued & CEPH_CAP_FILE_EXCL) {
2072                         ci->i_time_warp_seq++;
2073                         inode->i_atime = attr->ia_atime;
2074                         dirtied |= CEPH_CAP_FILE_EXCL;
2075                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2076                            timespec64_compare(&inode->i_atime,
2077                                             &attr->ia_atime) < 0) {
2078                         inode->i_atime = attr->ia_atime;
2079                         dirtied |= CEPH_CAP_FILE_WR;
2080                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2081                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2082                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2083                                                &attr->ia_atime);
2084                         mask |= CEPH_SETATTR_ATIME;
2085                         release |= CEPH_CAP_FILE_SHARED |
2086                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2087                 }
2088         }
2089         if (ia_valid & ATTR_MTIME) {
2090                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2091                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2092                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2093                 if (issued & CEPH_CAP_FILE_EXCL) {
2094                         ci->i_time_warp_seq++;
2095                         inode->i_mtime = attr->ia_mtime;
2096                         dirtied |= CEPH_CAP_FILE_EXCL;
2097                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2098                            timespec64_compare(&inode->i_mtime,
2099                                             &attr->ia_mtime) < 0) {
2100                         inode->i_mtime = attr->ia_mtime;
2101                         dirtied |= CEPH_CAP_FILE_WR;
2102                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2103                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2104                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2105                                                &attr->ia_mtime);
2106                         mask |= CEPH_SETATTR_MTIME;
2107                         release |= CEPH_CAP_FILE_SHARED |
2108                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2109                 }
2110         }
2111         if (ia_valid & ATTR_SIZE) {
2112                 dout("setattr %p size %lld -> %lld\n", inode,
2113                      inode->i_size, attr->ia_size);
2114                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2115                     attr->ia_size > inode->i_size) {
2116                         i_size_write(inode, attr->ia_size);
2117                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2118                         ci->i_reported_size = attr->ia_size;
2119                         dirtied |= CEPH_CAP_FILE_EXCL;
2120                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2121                            attr->ia_size != inode->i_size) {
2122                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2123                         req->r_args.setattr.old_size =
2124                                 cpu_to_le64(inode->i_size);
2125                         mask |= CEPH_SETATTR_SIZE;
2126                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2127                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2128                 }
2129         }
2130
2131         /* these do nothing */
2132         if (ia_valid & ATTR_CTIME) {
2133                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2134                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2135                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2136                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2137                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2138                      only ? "ctime only" : "ignored");
2139                 if (only) {
2140                         /*
2141                          * if kernel wants to dirty ctime but nothing else,
2142                          * we need to choose a cap to dirty under, or do
2143                          * a almost-no-op setattr
2144                          */
2145                         if (issued & CEPH_CAP_AUTH_EXCL)
2146                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2147                         else if (issued & CEPH_CAP_FILE_EXCL)
2148                                 dirtied |= CEPH_CAP_FILE_EXCL;
2149                         else if (issued & CEPH_CAP_XATTR_EXCL)
2150                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2151                         else
2152                                 mask |= CEPH_SETATTR_CTIME;
2153                 }
2154         }
2155         if (ia_valid & ATTR_FILE)
2156                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2157
2158         if (dirtied) {
2159                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2160                                                            &prealloc_cf);
2161                 inode->i_ctime = attr->ia_ctime;
2162         }
2163
2164         release &= issued;
2165         spin_unlock(&ci->i_ceph_lock);
2166         if (lock_snap_rwsem)
2167                 up_read(&mdsc->snap_rwsem);
2168
2169         if (inode_dirty_flags)
2170                 __mark_inode_dirty(inode, inode_dirty_flags);
2171
2172
2173         if (mask) {
2174                 req->r_inode = inode;
2175                 ihold(inode);
2176                 req->r_inode_drop = release;
2177                 req->r_args.setattr.mask = cpu_to_le32(mask);
2178                 req->r_num_caps = 1;
2179                 req->r_stamp = attr->ia_ctime;
2180                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2181         }
2182         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2183              ceph_cap_string(dirtied), mask);
2184
2185         ceph_mdsc_put_request(req);
2186         ceph_free_cap_flush(prealloc_cf);
2187
2188         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2189                 __ceph_do_pending_vmtruncate(inode);
2190
2191         return err;
2192 }
2193
2194 /*
2195  * setattr
2196  */
2197 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2198 {
2199         struct inode *inode = d_inode(dentry);
2200         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2201         int err;
2202
2203         if (ceph_snap(inode) != CEPH_NOSNAP)
2204                 return -EROFS;
2205
2206         err = setattr_prepare(dentry, attr);
2207         if (err != 0)
2208                 return err;
2209
2210         if ((attr->ia_valid & ATTR_SIZE) &&
2211             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2212                 return -EFBIG;
2213
2214         if ((attr->ia_valid & ATTR_SIZE) &&
2215             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2216                 return -EDQUOT;
2217
2218         err = __ceph_setattr(inode, attr);
2219
2220         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2221                 err = posix_acl_chmod(inode, attr->ia_mode);
2222
2223         return err;
2224 }
2225
2226 /*
2227  * Verify that we have a lease on the given mask.  If not,
2228  * do a getattr against an mds.
2229  */
2230 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2231                       int mask, bool force)
2232 {
2233         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2234         struct ceph_mds_client *mdsc = fsc->mdsc;
2235         struct ceph_mds_request *req;
2236         int mode;
2237         int err;
2238
2239         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2240                 dout("do_getattr inode %p SNAPDIR\n", inode);
2241                 return 0;
2242         }
2243
2244         dout("do_getattr inode %p mask %s mode 0%o\n",
2245              inode, ceph_cap_string(mask), inode->i_mode);
2246         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2247                 return 0;
2248
2249         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2250         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2251         if (IS_ERR(req))
2252                 return PTR_ERR(req);
2253         req->r_inode = inode;
2254         ihold(inode);
2255         req->r_num_caps = 1;
2256         req->r_args.getattr.mask = cpu_to_le32(mask);
2257         req->r_locked_page = locked_page;
2258         err = ceph_mdsc_do_request(mdsc, NULL, req);
2259         if (locked_page && err == 0) {
2260                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2261                 if (inline_version == 0) {
2262                         /* the reply is supposed to contain inline data */
2263                         err = -EINVAL;
2264                 } else if (inline_version == CEPH_INLINE_NONE) {
2265                         err = -ENODATA;
2266                 } else {
2267                         err = req->r_reply_info.targeti.inline_len;
2268                 }
2269         }
2270         ceph_mdsc_put_request(req);
2271         dout("do_getattr result=%d\n", err);
2272         return err;
2273 }
2274
2275
2276 /*
2277  * Check inode permissions.  We verify we have a valid value for
2278  * the AUTH cap, then call the generic handler.
2279  */
2280 int ceph_permission(struct inode *inode, int mask)
2281 {
2282         int err;
2283
2284         if (mask & MAY_NOT_BLOCK)
2285                 return -ECHILD;
2286
2287         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2288
2289         if (!err)
2290                 err = generic_permission(inode, mask);
2291         return err;
2292 }
2293
2294 /* Craft a mask of needed caps given a set of requested statx attrs. */
2295 static int statx_to_caps(u32 want)
2296 {
2297         int mask = 0;
2298
2299         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2300                 mask |= CEPH_CAP_AUTH_SHARED;
2301
2302         if (want & (STATX_NLINK|STATX_CTIME))
2303                 mask |= CEPH_CAP_LINK_SHARED;
2304
2305         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2306                     STATX_BLOCKS))
2307                 mask |= CEPH_CAP_FILE_SHARED;
2308
2309         if (want & (STATX_CTIME))
2310                 mask |= CEPH_CAP_XATTR_SHARED;
2311
2312         return mask;
2313 }
2314
2315 /*
2316  * Get all the attributes. If we have sufficient caps for the requested attrs,
2317  * then we can avoid talking to the MDS at all.
2318  */
2319 int ceph_getattr(const struct path *path, struct kstat *stat,
2320                  u32 request_mask, unsigned int flags)
2321 {
2322         struct inode *inode = d_inode(path->dentry);
2323         struct ceph_inode_info *ci = ceph_inode(inode);
2324         u32 valid_mask = STATX_BASIC_STATS;
2325         int err = 0;
2326
2327         /* Skip the getattr altogether if we're asked not to sync */
2328         if (!(flags & AT_STATX_DONT_SYNC)) {
2329                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2330                                       flags & AT_STATX_FORCE_SYNC);
2331                 if (err)
2332                         return err;
2333         }
2334
2335         generic_fillattr(inode, stat);
2336         stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2337
2338         /*
2339          * btime on newly-allocated inodes is 0, so if this is still set to
2340          * that, then assume that it's not valid.
2341          */
2342         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2343                 stat->btime = ci->i_btime;
2344                 valid_mask |= STATX_BTIME;
2345         }
2346
2347         if (ceph_snap(inode) == CEPH_NOSNAP)
2348                 stat->dev = inode->i_sb->s_dev;
2349         else
2350                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2351
2352         if (S_ISDIR(inode->i_mode)) {
2353                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2354                                         RBYTES))
2355                         stat->size = ci->i_rbytes;
2356                 else
2357                         stat->size = ci->i_files + ci->i_subdirs;
2358                 stat->blocks = 0;
2359                 stat->blksize = 65536;
2360                 /*
2361                  * Some applications rely on the number of st_nlink
2362                  * value on directories to be either 0 (if unlinked)
2363                  * or 2 + number of subdirectories.
2364                  */
2365                 if (stat->nlink == 1)
2366                         /* '.' + '..' + subdirs */
2367                         stat->nlink = 1 + 1 + ci->i_subdirs;
2368         }
2369
2370         stat->result_mask = request_mask & valid_mask;
2371         return err;
2372 }