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_buffer *old_blob = NULL;
 740         struct ceph_string *pool_ns = NULL;
 741         struct ceph_cap *new_cap = NULL;
 742         int err = 0;
 743         bool wake = false;
 744         bool queue_trunc = false;
 745         bool new_version = false;
 746         bool fill_inline = false;
 747
 748         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 749              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 750              ci->i_version);
 751
 752         info_caps = le32_to_cpu(info->cap.caps);
 753
 754         /* prealloc new cap struct */
 755         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
 756                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 757
 758         /*
 759          * prealloc xattr data, if it looks like we'll need it.  only
 760          * if len > 4 (meaning there are actually xattrs; the first 4
 761          * bytes are the xattr count).
 762          */
 763         if (iinfo->xattr_len > 4) {
 764                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 765                 if (!xattr_blob)
 766                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 767                                iinfo->xattr_len);
 768         }
 769
 770         if (iinfo->pool_ns_len > 0)
 771                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 772                                                      iinfo->pool_ns_len);
 773
 774         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 775                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 776
 777         spin_lock(&ci->i_ceph_lock);
 778
 779         /*
 780          * provided version will be odd if inode value is projected,
 781          * even if stable.  skip the update if we have newer stable
 782          * info (ours>=theirs, e.g. due to racing mds replies), unless
 783          * we are getting projected (unstable) info (in which case the
 784          * version is odd, and we want ours>theirs).
 785          *   us   them
 786          *   2    2     skip
 787          *   3    2     skip
 788          *   3    3     update
 789          */
 790         if (ci->i_version == 0 ||
 791             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 792              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 793                 new_version = true;
 794
 795         /* Update change_attribute */
 796         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 797
 798         __ceph_caps_issued(ci, &issued);
 799         issued |= __ceph_caps_dirty(ci);
 800         new_issued = ~issued & info_caps;
 801
 802         /* update inode */
 803         inode->i_rdev = le32_to_cpu(info->rdev);
 804         /* directories have fl_stripe_unit set to zero */
 805         if (le32_to_cpu(info->layout.fl_stripe_unit))
 806                 inode->i_blkbits =
 807                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 808         else
 809                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 810
 811         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 812
 813         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 814             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 815                 inode->i_mode = le32_to_cpu(info->mode);
 816                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 817                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 818                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 819                      from_kuid(&init_user_ns, inode->i_uid),
 820                      from_kgid(&init_user_ns, inode->i_gid));
 821                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 822                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 823         }
 824
 825         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 826             (issued & CEPH_CAP_LINK_EXCL) == 0)
 827                 set_nlink(inode, le32_to_cpu(info->nlink));
 828
 829         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 830                 /* be careful with mtime, atime, size */
 831                 ceph_decode_timespec64(&atime, &info->atime);
 832                 ceph_decode_timespec64(&mtime, &info->mtime);
 833                 ceph_decode_timespec64(&ctime, &info->ctime);
 834                 ceph_fill_file_time(inode, issued,
 835                                 le32_to_cpu(info->time_warp_seq),
 836                                 &ctime, &mtime, &atime);
 837         }
 838
 839         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 840                 ci->i_files = le64_to_cpu(info->files);
 841                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 842         }
 843
 844         if (new_version ||
 845             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 846                 s64 old_pool = ci->i_layout.pool_id;
 847                 struct ceph_string *old_ns;
 848
 849                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 850                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 851                                         lockdep_is_held(&ci->i_ceph_lock));
 852                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 853
 854                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 855                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 856
 857                 pool_ns = old_ns;
 858
 859                 queue_trunc = ceph_fill_file_size(inode, issued,
 860                                         le32_to_cpu(info->truncate_seq),
 861                                         le64_to_cpu(info->truncate_size),
 862                                         le64_to_cpu(info->size));
 863                 /* only update max_size on auth cap */
 864                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 865                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 866                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 867                                         le64_to_cpu(info->max_size));
 868                         ci->i_max_size = le64_to_cpu(info->max_size);
 869                 }
 870         }
 871
 872         /* layout and rstat are not tracked by capability, update them if
 873          * the inode info is from auth mds */
 874         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 875                 if (S_ISDIR(inode->i_mode)) {
 876                         ci->i_dir_layout = iinfo->dir_layout;
 877                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 878                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 879                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 880                         ci->i_dir_pin = iinfo->dir_pin;
 881                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 882                 }
 883         }
 884
 885         /* xattrs */
 886         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 887         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 888             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 889                 if (ci->i_xattrs.blob)
 890                         old_blob = ci->i_xattrs.blob;
 891                 ci->i_xattrs.blob = xattr_blob;
 892                 if (xattr_blob)
 893                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 894                                iinfo->xattr_data, iinfo->xattr_len);
 895                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 896                 ceph_forget_all_cached_acls(inode);
 897                 ceph_security_invalidate_secctx(inode);
 898                 xattr_blob = NULL;
 899         }
 900
 901         /* finally update i_version */
 902         if (le64_to_cpu(info->version) > ci->i_version)
 903                 ci->i_version = le64_to_cpu(info->version);
 904
 905         inode->i_mapping->a_ops = &ceph_aops;
 906
 907         switch (inode->i_mode & S_IFMT) {
 908         case S_IFIFO:
 909         case S_IFBLK:
 910         case S_IFCHR:
 911         case S_IFSOCK:
 912                 inode->i_blkbits = PAGE_SHIFT;
 913                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 914                 inode->i_op = &ceph_file_iops;
 915                 break;
 916         case S_IFREG:
 917                 inode->i_op = &ceph_file_iops;
 918                 inode->i_fop = &ceph_file_fops;
 919                 break;
 920         case S_IFLNK:
 921                 inode->i_op = &ceph_symlink_iops;
 922                 if (!ci->i_symlink) {
 923                         u32 symlen = iinfo->symlink_len;
 924                         char *sym;
 925
 926                         spin_unlock(&ci->i_ceph_lock);
 927
 928                         if (symlen != i_size_read(inode)) {
 929                                 pr_err("fill_inode %llx.%llx BAD symlink "
 930                                         "size %lld\n", ceph_vinop(inode),
 931                                         i_size_read(inode));
 932                                 i_size_write(inode, symlen);
 933                                 inode->i_blocks = calc_inode_blocks(symlen);
 934                         }
 935
 936                         err = -ENOMEM;
 937                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 938                         if (!sym)
 939                                 goto out;
 940
 941                         spin_lock(&ci->i_ceph_lock);
 942                         if (!ci->i_symlink)
 943                                 ci->i_symlink = sym;
 944                         else
 945                                 kfree(sym); /* lost a race */
 946                 }
 947                 inode->i_link = ci->i_symlink;
 948                 break;
 949         case S_IFDIR:
 950                 inode->i_op = &ceph_dir_iops;
 951                 inode->i_fop = &ceph_dir_fops;
 952                 break;
 953         default:
 954                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 955                        ceph_vinop(inode), inode->i_mode);
 956         }
 957
 958         /* were we issued a capability? */
 959         if (info_caps) {
 960                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 961                         ceph_add_cap(inode, session,
 962                                      le64_to_cpu(info->cap.cap_id),
 963                                      cap_fmode, info_caps,
 964                                      le32_to_cpu(info->cap.wanted),
 965                                      le32_to_cpu(info->cap.seq),
 966                                      le32_to_cpu(info->cap.mseq),
 967                                      le64_to_cpu(info->cap.realm),
 968                                      info->cap.flags, &new_cap);
 969
 970                         /* set dir completion flag? */
 971                         if (S_ISDIR(inode->i_mode) &&
 972                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 973                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 974                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 975                             !__ceph_dir_is_complete(ci)) {
 976                                 dout(" marking %p complete (empty)\n", inode);
 977                                 i_size_write(inode, 0);
 978                                 __ceph_dir_set_complete(ci,
 979                                         atomic64_read(&ci->i_release_count),
 980                                         atomic64_read(&ci->i_ordered_count));
 981                         }
 982
 983                         wake = true;
 984                 } else {
 985                         dout(" %p got snap_caps %s\n", inode,
 986                              ceph_cap_string(info_caps));
 987                         ci->i_snap_caps |= info_caps;
 988                         if (cap_fmode >= 0)
 989                                 __ceph_get_fmode(ci, cap_fmode);
 990                 }
 991         } else if (cap_fmode >= 0) {
 992                 pr_warn("mds issued no caps on %llx.%llx\n",
 993                            ceph_vinop(inode));
 994                 __ceph_get_fmode(ci, cap_fmode);
 995         }
 996
 997         if (iinfo->inline_version > 0 &&
 998             iinfo->inline_version >= ci->i_inline_version) {
 999                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1000                 ci->i_inline_version = iinfo->inline_version;
1001                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1002                     (locked_page || (info_caps & cache_caps)))
1003                         fill_inline = true;
1004         }
1005
1006         spin_unlock(&ci->i_ceph_lock);
1007
1008         if (fill_inline)
1009                 ceph_fill_inline_data(inode, locked_page,
1010                                       iinfo->inline_data, iinfo->inline_len);
1011
1012         if (wake)
1013                 wake_up_all(&ci->i_cap_wq);
1014
1015         /* queue truncate if we saw i_size decrease */
1016         if (queue_trunc)
1017                 ceph_queue_vmtruncate(inode);
1018
1019         /* populate frag tree */
1020         if (S_ISDIR(inode->i_mode))
1021                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1022
1023         /* update delegation info? */
1024         if (dirinfo)
1025                 ceph_fill_dirfrag(inode, dirinfo);
1026
1027         err = 0;
1028 out:
1029         if (new_cap)
1030                 ceph_put_cap(mdsc, new_cap);
1031         ceph_buffer_put(old_blob);
1032         ceph_buffer_put(xattr_blob);
1033         ceph_put_string(pool_ns);
1034         return err;
1035 }
1036
1037 /*
1038  * caller should hold session s_mutex and dentry->d_lock.
1039  */
1040 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1041                                   struct ceph_mds_reply_lease *lease,
1042                                   struct ceph_mds_session *session,
1043                                   unsigned long from_time,
1044                                   struct ceph_mds_session **old_lease_session)
1045 {
1046         struct ceph_dentry_info *di = ceph_dentry(dentry);
1047         long unsigned duration = le32_to_cpu(lease->duration_ms);
1048         long unsigned ttl = from_time + (duration * HZ) / 1000;
1049         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1050
1051         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1052              dentry, duration, ttl);
1053
1054         /* only track leases on regular dentries */
1055         if (ceph_snap(dir) != CEPH_NOSNAP)
1056                 return;
1057
1058         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1059         if (duration == 0) {
1060                 __ceph_dentry_dir_lease_touch(di);
1061                 return;
1062         }
1063
1064         if (di->lease_gen == session->s_cap_gen &&
1065             time_before(ttl, di->time))
1066                 return;  /* we already have a newer lease. */
1067
1068         if (di->lease_session && di->lease_session != session) {
1069                 *old_lease_session = di->lease_session;
1070                 di->lease_session = NULL;
1071         }
1072
1073         if (!di->lease_session)
1074                 di->lease_session = ceph_get_mds_session(session);
1075         di->lease_gen = session->s_cap_gen;
1076         di->lease_seq = le32_to_cpu(lease->seq);
1077         di->lease_renew_after = half_ttl;
1078         di->lease_renew_from = 0;
1079         di->time = ttl;
1080
1081         __ceph_dentry_lease_touch(di);
1082 }
1083
1084 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1085                                         struct ceph_mds_reply_lease *lease,
1086                                         struct ceph_mds_session *session,
1087                                         unsigned long from_time)
1088 {
1089         struct ceph_mds_session *old_lease_session = NULL;
1090         spin_lock(&dentry->d_lock);
1091         __update_dentry_lease(dir, dentry, lease, session, from_time,
1092                               &old_lease_session);
1093         spin_unlock(&dentry->d_lock);
1094         if (old_lease_session)
1095                 ceph_put_mds_session(old_lease_session);
1096 }
1097
1098 /*
1099  * update dentry lease without having parent inode locked
1100  */
1101 static void update_dentry_lease_careful(struct dentry *dentry,
1102                                         struct ceph_mds_reply_lease *lease,
1103                                         struct ceph_mds_session *session,
1104                                         unsigned long from_time,
1105                                         char *dname, u32 dname_len,
1106                                         struct ceph_vino *pdvino,
1107                                         struct ceph_vino *ptvino)
1108
1109 {
1110         struct inode *dir;
1111         struct ceph_mds_session *old_lease_session = NULL;
1112
1113         spin_lock(&dentry->d_lock);
1114         /* make sure dentry's name matches target */
1115         if (dentry->d_name.len != dname_len ||
1116             memcmp(dentry->d_name.name, dname, dname_len))
1117                 goto out_unlock;
1118
1119         dir = d_inode(dentry->d_parent);
1120         /* make sure parent matches dvino */
1121         if (!ceph_ino_compare(dir, pdvino))
1122                 goto out_unlock;
1123
1124         /* make sure dentry's inode matches target. NULL ptvino means that
1125          * we expect a negative dentry */
1126         if (ptvino) {
1127                 if (d_really_is_negative(dentry))
1128                         goto out_unlock;
1129                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1130                         goto out_unlock;
1131         } else {
1132                 if (d_really_is_positive(dentry))
1133                         goto out_unlock;
1134         }
1135
1136         __update_dentry_lease(dir, dentry, lease, session,
1137                               from_time, &old_lease_session);
1138 out_unlock:
1139         spin_unlock(&dentry->d_lock);
1140         if (old_lease_session)
1141                 ceph_put_mds_session(old_lease_session);
1142 }
1143
1144 /*
1145  * splice a dentry to an inode.
1146  * caller must hold directory i_mutex for this to be safe.
1147  */
1148 static int splice_dentry(struct dentry **pdn, struct inode *in)
1149 {
1150         struct dentry *dn = *pdn;
1151         struct dentry *realdn;
1152
1153         BUG_ON(d_inode(dn));
1154
1155         if (S_ISDIR(in->i_mode)) {
1156                 /* If inode is directory, d_splice_alias() below will remove
1157                  * 'realdn' from its origin parent. We need to ensure that
1158                  * origin parent's readdir cache will not reference 'realdn'
1159                  */
1160                 realdn = d_find_any_alias(in);
1161                 if (realdn) {
1162                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1163                         spin_lock(&realdn->d_lock);
1164
1165                         realdn->d_op->d_prune(realdn);
1166
1167                         di->time = jiffies;
1168                         di->lease_shared_gen = 0;
1169                         di->offset = 0;
1170
1171                         spin_unlock(&realdn->d_lock);
1172                         dput(realdn);
1173                 }
1174         }
1175
1176         /* dn must be unhashed */
1177         if (!d_unhashed(dn))
1178                 d_drop(dn);
1179         realdn = d_splice_alias(in, dn);
1180         if (IS_ERR(realdn)) {
1181                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1182                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1183                 return PTR_ERR(realdn);
1184         }
1185
1186         if (realdn) {
1187                 dout("dn %p (%d) spliced with %p (%d) "
1188                      "inode %p ino %llx.%llx\n",
1189                      dn, d_count(dn),
1190                      realdn, d_count(realdn),
1191                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1192                 dput(dn);
1193                 *pdn = realdn;
1194         } else {
1195                 BUG_ON(!ceph_dentry(dn));
1196                 dout("dn %p attached to %p ino %llx.%llx\n",
1197                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1198         }
1199         return 0;
1200 }
1201
1202 /*
1203  * Incorporate results into the local cache.  This is either just
1204  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1205  * after a lookup).
1206  *
1207  * A reply may contain
1208  *         a directory inode along with a dentry.
1209  *  and/or a target inode
1210  *
1211  * Called with snap_rwsem (read).
1212  */
1213 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1214 {
1215         struct ceph_mds_session *session = req->r_session;
1216         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1217         struct inode *in = NULL;
1218         struct ceph_vino tvino, dvino;
1219         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1220         int err = 0;
1221
1222         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1223              rinfo->head->is_dentry, rinfo->head->is_target);
1224
1225         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1226                 dout("fill_trace reply is empty!\n");
1227                 if (rinfo->head->result == 0 && req->r_parent)
1228                         ceph_invalidate_dir_request(req);
1229                 return 0;
1230         }
1231
1232         if (rinfo->head->is_dentry) {
1233                 struct inode *dir = req->r_parent;
1234
1235                 if (dir) {
1236                         err = fill_inode(dir, NULL,
1237                                          &rinfo->diri, rinfo->dirfrag,
1238                                          session, req->r_request_started, -1,
1239                                          &req->r_caps_reservation);
1240                         if (err < 0)
1241                                 goto done;
1242                 } else {
1243                         WARN_ON_ONCE(1);
1244                 }
1245
1246                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1247                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1248                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1249                         struct qstr dname;
1250                         struct dentry *dn, *parent;
1251
1252                         BUG_ON(!rinfo->head->is_target);
1253                         BUG_ON(req->r_dentry);
1254
1255                         parent = d_find_any_alias(dir);
1256                         BUG_ON(!parent);
1257
1258                         dname.name = rinfo->dname;
1259                         dname.len = rinfo->dname_len;
1260                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1261                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1262                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1263 retry_lookup:
1264                         dn = d_lookup(parent, &dname);
1265                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1266                              parent, dname.len, dname.name, dn);
1267
1268                         if (!dn) {
1269                                 dn = d_alloc(parent, &dname);
1270                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1271                                      dname.len, dname.name, dn);
1272                                 if (!dn) {
1273                                         dput(parent);
1274                                         err = -ENOMEM;
1275                                         goto done;
1276                                 }
1277                                 err = 0;
1278                         } else if (d_really_is_positive(dn) &&
1279                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1280                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1281                                 dout(" dn %p points to wrong inode %p\n",
1282                                      dn, d_inode(dn));
1283                                 ceph_dir_clear_ordered(dir);
1284                                 d_delete(dn);
1285                                 dput(dn);
1286                                 goto retry_lookup;
1287                         }
1288
1289                         req->r_dentry = dn;
1290                         dput(parent);
1291                 }
1292         }
1293
1294         if (rinfo->head->is_target) {
1295                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1296                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1297
1298                 in = ceph_get_inode(sb, tvino);
1299                 if (IS_ERR(in)) {
1300                         err = PTR_ERR(in);
1301                         goto done;
1302                 }
1303                 req->r_target_inode = in;
1304
1305                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1306                                 session, req->r_request_started,
1307                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1308                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1309                                 &req->r_caps_reservation);
1310                 if (err < 0) {
1311                         pr_err("fill_inode badness %p %llx.%llx\n",
1312                                 in, ceph_vinop(in));
1313                         goto done;
1314                 }
1315         }
1316
1317         /*
1318          * ignore null lease/binding on snapdir ENOENT, or else we
1319          * will have trouble splicing in the virtual snapdir later
1320          */
1321         if (rinfo->head->is_dentry &&
1322             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1323             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1324             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1325                                                fsc->mount_options->snapdir_name,
1326                                                req->r_dentry->d_name.len))) {
1327                 /*
1328                  * lookup link rename   : null -> possibly existing inode
1329                  * mknod symlink mkdir  : null -> new inode
1330                  * unlink               : linked -> null
1331                  */
1332                 struct inode *dir = req->r_parent;
1333                 struct dentry *dn = req->r_dentry;
1334                 bool have_dir_cap, have_lease;
1335
1336                 BUG_ON(!dn);
1337                 BUG_ON(!dir);
1338                 BUG_ON(d_inode(dn->d_parent) != dir);
1339
1340                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1341                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1342
1343                 BUG_ON(ceph_ino(dir) != dvino.ino);
1344                 BUG_ON(ceph_snap(dir) != dvino.snap);
1345
1346                 /* do we have a lease on the whole dir? */
1347                 have_dir_cap =
1348                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1349                          CEPH_CAP_FILE_SHARED);
1350
1351                 /* do we have a dn lease? */
1352                 have_lease = have_dir_cap ||
1353                         le32_to_cpu(rinfo->dlease->duration_ms);
1354                 if (!have_lease)
1355                         dout("fill_trace  no dentry lease or dir cap\n");
1356
1357                 /* rename? */
1358                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1359                         struct inode *olddir = req->r_old_dentry_dir;
1360                         BUG_ON(!olddir);
1361
1362                         dout(" src %p '%pd' dst %p '%pd'\n",
1363                              req->r_old_dentry,
1364                              req->r_old_dentry,
1365                              dn, dn);
1366                         dout("fill_trace doing d_move %p -> %p\n",
1367                              req->r_old_dentry, dn);
1368
1369                         /* d_move screws up sibling dentries' offsets */
1370                         ceph_dir_clear_ordered(dir);
1371                         ceph_dir_clear_ordered(olddir);
1372
1373                         d_move(req->r_old_dentry, dn);
1374                         dout(" src %p '%pd' dst %p '%pd'\n",
1375                              req->r_old_dentry,
1376                              req->r_old_dentry,
1377                              dn, dn);
1378
1379                         /* ensure target dentry is invalidated, despite
1380                            rehashing bug in vfs_rename_dir */
1381                         ceph_invalidate_dentry_lease(dn);
1382
1383                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1384                              ceph_dentry(req->r_old_dentry)->offset);
1385
1386                         /* swap r_dentry and r_old_dentry in case that
1387                          * splice_dentry() gets called later. This is safe
1388                          * because no other place will use them */
1389                         req->r_dentry = req->r_old_dentry;
1390                         req->r_old_dentry = dn;
1391                         dn = req->r_dentry;
1392                 }
1393
1394                 /* null dentry? */
1395                 if (!rinfo->head->is_target) {
1396                         dout("fill_trace null dentry\n");
1397                         if (d_really_is_positive(dn)) {
1398                                 dout("d_delete %p\n", dn);
1399                                 ceph_dir_clear_ordered(dir);
1400                                 d_delete(dn);
1401                         } else if (have_lease) {
1402                                 if (d_unhashed(dn))
1403                                         d_add(dn, NULL);
1404                                 update_dentry_lease(dir, dn,
1405                                                     rinfo->dlease, session,
1406                                                     req->r_request_started);
1407                         }
1408                         goto done;
1409                 }
1410
1411                 /* attach proper inode */
1412                 if (d_really_is_negative(dn)) {
1413                         ceph_dir_clear_ordered(dir);
1414                         ihold(in);
1415                         err = splice_dentry(&req->r_dentry, in);
1416                         if (err < 0)
1417                                 goto done;
1418                         dn = req->r_dentry;  /* may have spliced */
1419                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1420                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1421                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1422                              ceph_vinop(in));
1423                         d_invalidate(dn);
1424                         have_lease = false;
1425                 }
1426
1427                 if (have_lease) {
1428                         update_dentry_lease(dir, dn,
1429                                             rinfo->dlease, session,
1430                                             req->r_request_started);
1431                 }
1432                 dout(" final dn %p\n", dn);
1433         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1434                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1435                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1436                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1437                 struct inode *dir = req->r_parent;
1438
1439                 /* fill out a snapdir LOOKUPSNAP dentry */
1440                 BUG_ON(!dir);
1441                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1442                 BUG_ON(!req->r_dentry);
1443                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1444                 ceph_dir_clear_ordered(dir);
1445                 ihold(in);
1446                 err = splice_dentry(&req->r_dentry, in);
1447                 if (err < 0)
1448                         goto done;
1449         } else if (rinfo->head->is_dentry && req->r_dentry) {
1450                 /* parent inode is not locked, be carefull */
1451                 struct ceph_vino *ptvino = NULL;
1452                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1453                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1454                 if (rinfo->head->is_target) {
1455                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1456                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1457                         ptvino = &tvino;
1458                 }
1459                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1460                                             session, req->r_request_started,
1461                                             rinfo->dname, rinfo->dname_len,
1462                                             &dvino, ptvino);
1463         }
1464 done:
1465         dout("fill_trace done err=%d\n", err);
1466         return err;
1467 }
1468
1469 /*
1470  * Prepopulate our cache with readdir results, leases, etc.
1471  */
1472 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1473                                            struct ceph_mds_session *session)
1474 {
1475         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1476         int i, err = 0;
1477
1478         for (i = 0; i < rinfo->dir_nr; i++) {
1479                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1480                 struct ceph_vino vino;
1481                 struct inode *in;
1482                 int rc;
1483
1484                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1485                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1486
1487                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1488                 if (IS_ERR(in)) {
1489                         err = PTR_ERR(in);
1490                         dout("new_inode badness got %d\n", err);
1491                         continue;
1492                 }
1493                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1494                                 req->r_request_started, -1,
1495                                 &req->r_caps_reservation);
1496                 if (rc < 0) {
1497                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1498                         err = rc;
1499                 }
1500                 /* avoid calling iput_final() in mds dispatch threads */
1501                 ceph_async_iput(in);
1502         }
1503
1504         return err;
1505 }
1506
1507 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1508 {
1509         if (ctl->page) {
1510                 kunmap(ctl->page);
1511                 put_page(ctl->page);
1512                 ctl->page = NULL;
1513         }
1514 }
1515
1516 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1517                               struct ceph_readdir_cache_control *ctl,
1518                               struct ceph_mds_request *req)
1519 {
1520         struct ceph_inode_info *ci = ceph_inode(dir);
1521         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1522         unsigned idx = ctl->index % nsize;
1523         pgoff_t pgoff = ctl->index / nsize;
1524
1525         if (!ctl->page || pgoff != page_index(ctl->page)) {
1526                 ceph_readdir_cache_release(ctl);
1527                 if (idx == 0)
1528                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1529                 else
1530                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1531                 if (!ctl->page) {
1532                         ctl->index = -1;
1533                         return idx == 0 ? -ENOMEM : 0;
1534                 }
1535                 /* reading/filling the cache are serialized by
1536                  * i_mutex, no need to use page lock */
1537                 unlock_page(ctl->page);
1538                 ctl->dentries = kmap(ctl->page);
1539                 if (idx == 0)
1540                         memset(ctl->dentries, 0, PAGE_SIZE);
1541         }
1542
1543         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1544             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1545                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1546                 ctl->dentries[idx] = dn;
1547                 ctl->index++;
1548         } else {
1549                 dout("disable readdir cache\n");
1550                 ctl->index = -1;
1551         }
1552         return 0;
1553 }
1554
1555 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1556                              struct ceph_mds_session *session)
1557 {
1558         struct dentry *parent = req->r_dentry;
1559         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1560         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1561         struct qstr dname;
1562         struct dentry *dn;
1563         struct inode *in;
1564         int err = 0, skipped = 0, ret, i;
1565         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1566         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1567         u32 last_hash = 0;
1568         u32 fpos_offset;
1569         struct ceph_readdir_cache_control cache_ctl = {};
1570
1571         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1572                 return readdir_prepopulate_inodes_only(req, session);
1573
1574         if (rinfo->hash_order) {
1575                 if (req->r_path2) {
1576                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1577                                                   req->r_path2,
1578                                                   strlen(req->r_path2));
1579                         last_hash = ceph_frag_value(last_hash);
1580                 } else if (rinfo->offset_hash) {
1581                         /* mds understands offset_hash */
1582                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1583                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1584                 }
1585         }
1586
1587         if (rinfo->dir_dir &&
1588             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1589                 dout("readdir_prepopulate got new frag %x -> %x\n",
1590                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1591                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1592                 if (!rinfo->hash_order)
1593                         req->r_readdir_offset = 2;
1594         }
1595
1596         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1597                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1598                      rinfo->dir_nr, parent);
1599         } else {
1600                 dout("readdir_prepopulate %d items under dn %p\n",
1601                      rinfo->dir_nr, parent);
1602                 if (rinfo->dir_dir)
1603                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1604
1605                 if (ceph_frag_is_leftmost(frag) &&
1606                     req->r_readdir_offset == 2 &&
1607                     !(rinfo->hash_order && last_hash)) {
1608                         /* note dir version at start of readdir so we can
1609                          * tell if any dentries get dropped */
1610                         req->r_dir_release_cnt =
1611                                 atomic64_read(&ci->i_release_count);
1612                         req->r_dir_ordered_cnt =
1613                                 atomic64_read(&ci->i_ordered_count);
1614                         req->r_readdir_cache_idx = 0;
1615                 }
1616         }
1617
1618         cache_ctl.index = req->r_readdir_cache_idx;
1619         fpos_offset = req->r_readdir_offset;
1620
1621         /* FIXME: release caps/leases if error occurs */
1622         for (i = 0; i < rinfo->dir_nr; i++) {
1623                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1624                 struct ceph_vino tvino;
1625
1626                 dname.name = rde->name;
1627                 dname.len = rde->name_len;
1628                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1629
1630                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1631                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1632
1633                 if (rinfo->hash_order) {
1634                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1635                                                  rde->name, rde->name_len);
1636                         hash = ceph_frag_value(hash);
1637                         if (hash != last_hash)
1638                                 fpos_offset = 2;
1639                         last_hash = hash;
1640                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1641                 } else {
1642                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1643                 }
1644
1645 retry_lookup:
1646                 dn = d_lookup(parent, &dname);
1647                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1648                      parent, dname.len, dname.name, dn);
1649
1650                 if (!dn) {
1651                         dn = d_alloc(parent, &dname);
1652                         dout("d_alloc %p '%.*s' = %p\n", parent,
1653                              dname.len, dname.name, dn);
1654                         if (!dn) {
1655                                 dout("d_alloc badness\n");
1656                                 err = -ENOMEM;
1657                                 goto out;
1658                         }
1659                 } else if (d_really_is_positive(dn) &&
1660                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1661                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1662                         struct ceph_dentry_info *di = ceph_dentry(dn);
1663                         dout(" dn %p points to wrong inode %p\n",
1664                              dn, d_inode(dn));
1665
1666                         spin_lock(&dn->d_lock);
1667                         if (di->offset > 0 &&
1668                             di->lease_shared_gen ==
1669                             atomic_read(&ci->i_shared_gen)) {
1670                                 __ceph_dir_clear_ordered(ci);
1671                                 di->offset = 0;
1672                         }
1673                         spin_unlock(&dn->d_lock);
1674
1675                         d_delete(dn);
1676                         dput(dn);
1677                         goto retry_lookup;
1678                 }
1679
1680                 /* inode */
1681                 if (d_really_is_positive(dn)) {
1682                         in = d_inode(dn);
1683                 } else {
1684                         in = ceph_get_inode(parent->d_sb, tvino);
1685                         if (IS_ERR(in)) {
1686                                 dout("new_inode badness\n");
1687                                 d_drop(dn);
1688                                 dput(dn);
1689                                 err = PTR_ERR(in);
1690                                 goto out;
1691                         }
1692                 }
1693
1694                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1695                                  req->r_request_started, -1,
1696                                  &req->r_caps_reservation);
1697                 if (ret < 0) {
1698                         pr_err("fill_inode badness on %p\n", in);
1699                         if (d_really_is_negative(dn)) {
1700                                 /* avoid calling iput_final() in mds
1701                                  * dispatch threads */
1702                                 ceph_async_iput(in);
1703                         }
1704                         d_drop(dn);
1705                         err = ret;
1706                         goto next_item;
1707                 }
1708
1709                 if (d_really_is_negative(dn)) {
1710                         if (ceph_security_xattr_deadlock(in)) {
1711                                 dout(" skip splicing dn %p to inode %p"
1712                                      " (security xattr deadlock)\n", dn, in);
1713                                 ceph_async_iput(in);
1714                                 skipped++;
1715                                 goto next_item;
1716                         }
1717
1718                         err = splice_dentry(&dn, in);
1719                         if (err < 0)
1720                                 goto next_item;
1721                 }
1722
1723                 ceph_dentry(dn)->offset = rde->offset;
1724
1725                 update_dentry_lease(d_inode(parent), dn,
1726                                     rde->lease, req->r_session,
1727                                     req->r_request_started);
1728
1729                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1730                         ret = fill_readdir_cache(d_inode(parent), dn,
1731                                                  &cache_ctl, req);
1732                         if (ret < 0)
1733                                 err = ret;
1734                 }
1735 next_item:
1736                 dput(dn);
1737         }
1738 out:
1739         if (err == 0 && skipped == 0) {
1740                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1741                 req->r_readdir_cache_idx = cache_ctl.index;
1742         }
1743         ceph_readdir_cache_release(&cache_ctl);
1744         dout("readdir_prepopulate done\n");
1745         return err;
1746 }
1747
1748 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1749 {
1750         struct ceph_inode_info *ci = ceph_inode(inode);
1751         bool ret;
1752
1753         spin_lock(&ci->i_ceph_lock);
1754         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1755         i_size_write(inode, size);
1756         inode->i_blocks = calc_inode_blocks(size);
1757
1758         ret = __ceph_should_report_size(ci);
1759
1760         spin_unlock(&ci->i_ceph_lock);
1761         return ret;
1762 }
1763
1764 /*
1765  * Put reference to inode, but avoid calling iput_final() in current thread.
1766  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1767  * some contexts.
1768  */
1769 void ceph_async_iput(struct inode *inode)
1770 {
1771         if (!inode)
1772                 return;
1773         for (;;) {
1774                 if (atomic_add_unless(&inode->i_count, -1, 1))
1775                         break;
1776                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1777                                &ceph_inode(inode)->i_work))
1778                         break;
1779                 /* queue work failed, i_count must be at least 2 */
1780         }
1781 }
1782
1783 /*
1784  * Write back inode data in a worker thread.  (This can't be done
1785  * in the message handler context.)
1786  */
1787 void ceph_queue_writeback(struct inode *inode)
1788 {
1789         struct ceph_inode_info *ci = ceph_inode(inode);
1790         set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1791
1792         ihold(inode);
1793         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1794                        &ci->i_work)) {
1795                 dout("ceph_queue_writeback %p\n", inode);
1796         } else {
1797                 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1798                      inode, ci->i_work_mask);
1799                 iput(inode);
1800         }
1801 }
1802
1803 /*
1804  * queue an async invalidation
1805  */
1806 void ceph_queue_invalidate(struct inode *inode)
1807 {
1808         struct ceph_inode_info *ci = ceph_inode(inode);
1809         set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1810
1811         ihold(inode);
1812         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1813                        &ceph_inode(inode)->i_work)) {
1814                 dout("ceph_queue_invalidate %p\n", inode);
1815         } else {
1816                 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1817                      inode, ci->i_work_mask);
1818                 iput(inode);
1819         }
1820 }
1821
1822 /*
1823  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1824  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1825  */
1826 void ceph_queue_vmtruncate(struct inode *inode)
1827 {
1828         struct ceph_inode_info *ci = ceph_inode(inode);
1829         set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1830
1831         ihold(inode);
1832         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1833                        &ci->i_work)) {
1834                 dout("ceph_queue_vmtruncate %p\n", inode);
1835         } else {
1836                 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1837                      inode, ci->i_work_mask);
1838                 iput(inode);
1839         }
1840 }
1841
1842 static void ceph_do_invalidate_pages(struct inode *inode)
1843 {
1844         struct ceph_inode_info *ci = ceph_inode(inode);
1845         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1846         u32 orig_gen;
1847         int check = 0;
1848
1849         mutex_lock(&ci->i_truncate_mutex);
1850
1851         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1852                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1853                                     inode, ceph_ino(inode));
1854                 mapping_set_error(inode->i_mapping, -EIO);
1855                 truncate_pagecache(inode, 0);
1856                 mutex_unlock(&ci->i_truncate_mutex);
1857                 goto out;
1858         }
1859
1860         spin_lock(&ci->i_ceph_lock);
1861         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1862              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1863         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1864                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1865                         check = 1;
1866                 spin_unlock(&ci->i_ceph_lock);
1867                 mutex_unlock(&ci->i_truncate_mutex);
1868                 goto out;
1869         }
1870         orig_gen = ci->i_rdcache_gen;
1871         spin_unlock(&ci->i_ceph_lock);
1872
1873         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1874                 pr_err("invalidate_pages %p fails\n", inode);
1875         }
1876
1877         spin_lock(&ci->i_ceph_lock);
1878         if (orig_gen == ci->i_rdcache_gen &&
1879             orig_gen == ci->i_rdcache_revoking) {
1880                 dout("invalidate_pages %p gen %d successful\n", inode,
1881                      ci->i_rdcache_gen);
1882                 ci->i_rdcache_revoking--;
1883                 check = 1;
1884         } else {
1885                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1886                      inode, orig_gen, ci->i_rdcache_gen,
1887                      ci->i_rdcache_revoking);
1888                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1889                         check = 1;
1890         }
1891         spin_unlock(&ci->i_ceph_lock);
1892         mutex_unlock(&ci->i_truncate_mutex);
1893 out:
1894         if (check)
1895                 ceph_check_caps(ci, 0, NULL);
1896 }
1897
1898 /*
1899  * Make sure any pending truncation is applied before doing anything
1900  * that may depend on it.
1901  */
1902 void __ceph_do_pending_vmtruncate(struct inode *inode)
1903 {
1904         struct ceph_inode_info *ci = ceph_inode(inode);
1905         u64 to;
1906         int wrbuffer_refs, finish = 0;
1907
1908         mutex_lock(&ci->i_truncate_mutex);
1909 retry:
1910         spin_lock(&ci->i_ceph_lock);
1911         if (ci->i_truncate_pending == 0) {
1912                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1913                 spin_unlock(&ci->i_ceph_lock);
1914                 mutex_unlock(&ci->i_truncate_mutex);
1915                 return;
1916         }
1917
1918         /*
1919          * make sure any dirty snapped pages are flushed before we
1920          * possibly truncate them.. so write AND block!
1921          */
1922         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1923                 spin_unlock(&ci->i_ceph_lock);
1924                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1925                      inode);
1926                 filemap_write_and_wait_range(&inode->i_data, 0,
1927                                              inode->i_sb->s_maxbytes);
1928                 goto retry;
1929         }
1930
1931         /* there should be no reader or writer */
1932         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1933
1934         to = ci->i_truncate_size;
1935         wrbuffer_refs = ci->i_wrbuffer_ref;
1936         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1937              ci->i_truncate_pending, to);
1938         spin_unlock(&ci->i_ceph_lock);
1939
1940         truncate_pagecache(inode, to);
1941
1942         spin_lock(&ci->i_ceph_lock);
1943         if (to == ci->i_truncate_size) {
1944                 ci->i_truncate_pending = 0;
1945                 finish = 1;
1946         }
1947         spin_unlock(&ci->i_ceph_lock);
1948         if (!finish)
1949                 goto retry;
1950
1951         mutex_unlock(&ci->i_truncate_mutex);
1952
1953         if (wrbuffer_refs == 0)
1954                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1955
1956         wake_up_all(&ci->i_cap_wq);
1957 }
1958
1959 static void ceph_inode_work(struct work_struct *work)
1960 {
1961         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1962                                                  i_work);
1963         struct inode *inode = &ci->vfs_inode;
1964
1965         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1966                 dout("writeback %p\n", inode);
1967                 filemap_fdatawrite(&inode->i_data);
1968         }
1969         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1970                 ceph_do_invalidate_pages(inode);
1971
1972         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1973                 __ceph_do_pending_vmtruncate(inode);
1974
1975         iput(inode);
1976 }
1977
1978 /*
1979  * symlinks
1980  */
1981 static const struct inode_operations ceph_symlink_iops = {
1982         .get_link = simple_get_link,
1983         .setattr = ceph_setattr,
1984         .getattr = ceph_getattr,
1985         .listxattr = ceph_listxattr,
1986 };
1987
1988 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1989 {
1990         struct ceph_inode_info *ci = ceph_inode(inode);
1991         const unsigned int ia_valid = attr->ia_valid;
1992         struct ceph_mds_request *req;
1993         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1994         struct ceph_cap_flush *prealloc_cf;
1995         int issued;
1996         int release = 0, dirtied = 0;
1997         int mask = 0;
1998         int err = 0;
1999         int inode_dirty_flags = 0;
2000         bool lock_snap_rwsem = false;
2001
2002         prealloc_cf = ceph_alloc_cap_flush();
2003         if (!prealloc_cf)
2004                 return -ENOMEM;
2005
2006         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2007                                        USE_AUTH_MDS);
2008         if (IS_ERR(req)) {
2009                 ceph_free_cap_flush(prealloc_cf);
2010                 return PTR_ERR(req);
2011         }
2012
2013         spin_lock(&ci->i_ceph_lock);
2014         issued = __ceph_caps_issued(ci, NULL);
2015
2016         if (!ci->i_head_snapc &&
2017             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2018                 lock_snap_rwsem = true;
2019                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2020                         spin_unlock(&ci->i_ceph_lock);
2021                         down_read(&mdsc->snap_rwsem);
2022                         spin_lock(&ci->i_ceph_lock);
2023                         issued = __ceph_caps_issued(ci, NULL);
2024                 }
2025         }
2026
2027         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2028
2029         if (ia_valid & ATTR_UID) {
2030                 dout("setattr %p uid %d -> %d\n", inode,
2031                      from_kuid(&init_user_ns, inode->i_uid),
2032                      from_kuid(&init_user_ns, attr->ia_uid));
2033                 if (issued & CEPH_CAP_AUTH_EXCL) {
2034                         inode->i_uid = attr->ia_uid;
2035                         dirtied |= CEPH_CAP_AUTH_EXCL;
2036                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2037                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2038                         req->r_args.setattr.uid = cpu_to_le32(
2039                                 from_kuid(&init_user_ns, attr->ia_uid));
2040                         mask |= CEPH_SETATTR_UID;
2041                         release |= CEPH_CAP_AUTH_SHARED;
2042                 }
2043         }
2044         if (ia_valid & ATTR_GID) {
2045                 dout("setattr %p gid %d -> %d\n", inode,
2046                      from_kgid(&init_user_ns, inode->i_gid),
2047                      from_kgid(&init_user_ns, attr->ia_gid));
2048                 if (issued & CEPH_CAP_AUTH_EXCL) {
2049                         inode->i_gid = attr->ia_gid;
2050                         dirtied |= CEPH_CAP_AUTH_EXCL;
2051                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2052                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2053                         req->r_args.setattr.gid = cpu_to_le32(
2054                                 from_kgid(&init_user_ns, attr->ia_gid));
2055                         mask |= CEPH_SETATTR_GID;
2056                         release |= CEPH_CAP_AUTH_SHARED;
2057                 }
2058         }
2059         if (ia_valid & ATTR_MODE) {
2060                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2061                      attr->ia_mode);
2062                 if (issued & CEPH_CAP_AUTH_EXCL) {
2063                         inode->i_mode = attr->ia_mode;
2064                         dirtied |= CEPH_CAP_AUTH_EXCL;
2065                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2066                            attr->ia_mode != inode->i_mode) {
2067                         inode->i_mode = attr->ia_mode;
2068                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2069                         mask |= CEPH_SETATTR_MODE;
2070                         release |= CEPH_CAP_AUTH_SHARED;
2071                 }
2072         }
2073
2074         if (ia_valid & ATTR_ATIME) {
2075                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2076                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2077                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2078                 if (issued & CEPH_CAP_FILE_EXCL) {
2079                         ci->i_time_warp_seq++;
2080                         inode->i_atime = attr->ia_atime;
2081                         dirtied |= CEPH_CAP_FILE_EXCL;
2082                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2083                            timespec64_compare(&inode->i_atime,
2084                                             &attr->ia_atime) < 0) {
2085                         inode->i_atime = attr->ia_atime;
2086                         dirtied |= CEPH_CAP_FILE_WR;
2087                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2088                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2089                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2090                                                &attr->ia_atime);
2091                         mask |= CEPH_SETATTR_ATIME;
2092                         release |= CEPH_CAP_FILE_SHARED |
2093                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2094                 }
2095         }
2096         if (ia_valid & ATTR_MTIME) {
2097                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2098                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2099                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2100                 if (issued & CEPH_CAP_FILE_EXCL) {
2101                         ci->i_time_warp_seq++;
2102                         inode->i_mtime = attr->ia_mtime;
2103                         dirtied |= CEPH_CAP_FILE_EXCL;
2104                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2105                            timespec64_compare(&inode->i_mtime,
2106                                             &attr->ia_mtime) < 0) {
2107                         inode->i_mtime = attr->ia_mtime;
2108                         dirtied |= CEPH_CAP_FILE_WR;
2109                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2110                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2111                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2112                                                &attr->ia_mtime);
2113                         mask |= CEPH_SETATTR_MTIME;
2114                         release |= CEPH_CAP_FILE_SHARED |
2115                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2116                 }
2117         }
2118         if (ia_valid & ATTR_SIZE) {
2119                 dout("setattr %p size %lld -> %lld\n", inode,
2120                      inode->i_size, attr->ia_size);
2121                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2122                     attr->ia_size > inode->i_size) {
2123                         i_size_write(inode, attr->ia_size);
2124                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2125                         ci->i_reported_size = attr->ia_size;
2126                         dirtied |= CEPH_CAP_FILE_EXCL;
2127                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2128                            attr->ia_size != inode->i_size) {
2129                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2130                         req->r_args.setattr.old_size =
2131                                 cpu_to_le64(inode->i_size);
2132                         mask |= CEPH_SETATTR_SIZE;
2133                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2134                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2135                 }
2136         }
2137
2138         /* these do nothing */
2139         if (ia_valid & ATTR_CTIME) {
2140                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2141                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2142                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2143                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2144                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2145                      only ? "ctime only" : "ignored");
2146                 if (only) {
2147                         /*
2148                          * if kernel wants to dirty ctime but nothing else,
2149                          * we need to choose a cap to dirty under, or do
2150                          * a almost-no-op setattr
2151                          */
2152                         if (issued & CEPH_CAP_AUTH_EXCL)
2153                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2154                         else if (issued & CEPH_CAP_FILE_EXCL)
2155                                 dirtied |= CEPH_CAP_FILE_EXCL;
2156                         else if (issued & CEPH_CAP_XATTR_EXCL)
2157                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2158                         else
2159                                 mask |= CEPH_SETATTR_CTIME;
2160                 }
2161         }
2162         if (ia_valid & ATTR_FILE)
2163                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2164
2165         if (dirtied) {
2166                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2167                                                            &prealloc_cf);
2168                 inode->i_ctime = attr->ia_ctime;
2169         }
2170
2171         release &= issued;
2172         spin_unlock(&ci->i_ceph_lock);
2173         if (lock_snap_rwsem)
2174                 up_read(&mdsc->snap_rwsem);
2175
2176         if (inode_dirty_flags)
2177                 __mark_inode_dirty(inode, inode_dirty_flags);
2178
2179
2180         if (mask) {
2181                 req->r_inode = inode;
2182                 ihold(inode);
2183                 req->r_inode_drop = release;
2184                 req->r_args.setattr.mask = cpu_to_le32(mask);
2185                 req->r_num_caps = 1;
2186                 req->r_stamp = attr->ia_ctime;
2187                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2188         }
2189         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2190              ceph_cap_string(dirtied), mask);
2191
2192         ceph_mdsc_put_request(req);
2193         ceph_free_cap_flush(prealloc_cf);
2194
2195         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2196                 __ceph_do_pending_vmtruncate(inode);
2197
2198         return err;
2199 }
2200
2201 /*
2202  * setattr
2203  */
2204 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2205 {
2206         struct inode *inode = d_inode(dentry);
2207         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2208         int err;
2209
2210         if (ceph_snap(inode) != CEPH_NOSNAP)
2211                 return -EROFS;
2212
2213         err = setattr_prepare(dentry, attr);
2214         if (err != 0)
2215                 return err;
2216
2217         if ((attr->ia_valid & ATTR_SIZE) &&
2218             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2219                 return -EFBIG;
2220
2221         if ((attr->ia_valid & ATTR_SIZE) &&
2222             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2223                 return -EDQUOT;
2224
2225         err = __ceph_setattr(inode, attr);
2226
2227         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2228                 err = posix_acl_chmod(inode, attr->ia_mode);
2229
2230         return err;
2231 }
2232
2233 /*
2234  * Verify that we have a lease on the given mask.  If not,
2235  * do a getattr against an mds.
2236  */
2237 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2238                       int mask, bool force)
2239 {
2240         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2241         struct ceph_mds_client *mdsc = fsc->mdsc;
2242         struct ceph_mds_request *req;
2243         int mode;
2244         int err;
2245
2246         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2247                 dout("do_getattr inode %p SNAPDIR\n", inode);
2248                 return 0;
2249         }
2250
2251         dout("do_getattr inode %p mask %s mode 0%o\n",
2252              inode, ceph_cap_string(mask), inode->i_mode);
2253         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2254                 return 0;
2255
2256         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2257         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2258         if (IS_ERR(req))
2259                 return PTR_ERR(req);
2260         req->r_inode = inode;
2261         ihold(inode);
2262         req->r_num_caps = 1;
2263         req->r_args.getattr.mask = cpu_to_le32(mask);
2264         req->r_locked_page = locked_page;
2265         err = ceph_mdsc_do_request(mdsc, NULL, req);
2266         if (locked_page && err == 0) {
2267                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2268                 if (inline_version == 0) {
2269                         /* the reply is supposed to contain inline data */
2270                         err = -EINVAL;
2271                 } else if (inline_version == CEPH_INLINE_NONE) {
2272                         err = -ENODATA;
2273                 } else {
2274                         err = req->r_reply_info.targeti.inline_len;
2275                 }
2276         }
2277         ceph_mdsc_put_request(req);
2278         dout("do_getattr result=%d\n", err);
2279         return err;
2280 }
2281
2282
2283 /*
2284  * Check inode permissions.  We verify we have a valid value for
2285  * the AUTH cap, then call the generic handler.
2286  */
2287 int ceph_permission(struct inode *inode, int mask)
2288 {
2289         int err;
2290
2291         if (mask & MAY_NOT_BLOCK)
2292                 return -ECHILD;
2293
2294         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2295
2296         if (!err)
2297                 err = generic_permission(inode, mask);
2298         return err;
2299 }
2300
2301 /* Craft a mask of needed caps given a set of requested statx attrs. */
2302 static int statx_to_caps(u32 want)
2303 {
2304         int mask = 0;
2305
2306         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2307                 mask |= CEPH_CAP_AUTH_SHARED;
2308
2309         if (want & (STATX_NLINK|STATX_CTIME))
2310                 mask |= CEPH_CAP_LINK_SHARED;
2311
2312         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2313                     STATX_BLOCKS))
2314                 mask |= CEPH_CAP_FILE_SHARED;
2315
2316         if (want & (STATX_CTIME))
2317                 mask |= CEPH_CAP_XATTR_SHARED;
2318
2319         return mask;
2320 }
2321
2322 /*
2323  * Get all the attributes. If we have sufficient caps for the requested attrs,
2324  * then we can avoid talking to the MDS at all.
2325  */
2326 int ceph_getattr(const struct path *path, struct kstat *stat,
2327                  u32 request_mask, unsigned int flags)
2328 {
2329         struct inode *inode = d_inode(path->dentry);
2330         struct ceph_inode_info *ci = ceph_inode(inode);
2331         u32 valid_mask = STATX_BASIC_STATS;
2332         int err = 0;
2333
2334         /* Skip the getattr altogether if we're asked not to sync */
2335         if (!(flags & AT_STATX_DONT_SYNC)) {
2336                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2337                                       flags & AT_STATX_FORCE_SYNC);
2338                 if (err)
2339                         return err;
2340         }
2341
2342         generic_fillattr(inode, stat);
2343         stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2344
2345         /*
2346          * btime on newly-allocated inodes is 0, so if this is still set to
2347          * that, then assume that it's not valid.
2348          */
2349         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2350                 stat->btime = ci->i_btime;
2351                 valid_mask |= STATX_BTIME;
2352         }
2353
2354         if (ceph_snap(inode) == CEPH_NOSNAP)
2355                 stat->dev = inode->i_sb->s_dev;
2356         else
2357                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2358
2359         if (S_ISDIR(inode->i_mode)) {
2360                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2361                                         RBYTES))
2362                         stat->size = ci->i_rbytes;
2363                 else
2364                         stat->size = ci->i_files + ci->i_subdirs;
2365                 stat->blocks = 0;
2366                 stat->blksize = 65536;
2367                 /*
2368                  * Some applications rely on the number of st_nlink
2369                  * value on directories to be either 0 (if unlinked)
2370                  * or 2 + number of subdirectories.
2371                  */
2372                 if (stat->nlink == 1)
2373                         /* '.' + '..' + subdirs */
2374                         stat->nlink = 1 + 1 + ci->i_subdirs;
2375         }
2376
2377         stat->result_mask = request_mask & valid_mask;
2378         return err;
2379 }