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         struct ceph_inode_info *ci = ceph_inode(inode);
  45
  46         ci->i_vino = *(struct ceph_vino *)data;
  47         inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
  48         inode_set_iversion_raw(inode, 0);
  49         return 0;
  50 }
  51
  52 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  53 {
  54         struct inode *inode;
  55
  56         inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
  57                              ceph_set_ino_cb, &vino);
  58         if (!inode)
  59                 return ERR_PTR(-ENOMEM);
  60
  61         dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
  62              ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
  63         return inode;
  64 }
  65
  66 /*
  67  * get/constuct snapdir inode for a given directory
  68  */
  69 struct inode *ceph_get_snapdir(struct inode *parent)
  70 {
  71         struct ceph_vino vino = {
  72                 .ino = ceph_ino(parent),
  73                 .snap = CEPH_SNAPDIR,
  74         };
  75         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  76         struct ceph_inode_info *ci = ceph_inode(inode);
  77
  78         BUG_ON(!S_ISDIR(parent->i_mode));
  79         if (IS_ERR(inode))
  80                 return inode;
  81         inode->i_mode = parent->i_mode;
  82         inode->i_uid = parent->i_uid;
  83         inode->i_gid = parent->i_gid;
  84         inode->i_mtime = parent->i_mtime;
  85         inode->i_ctime = parent->i_ctime;
  86         inode->i_atime = parent->i_atime;
  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         ci->i_btime = ceph_inode(parent)->i_btime;
  92
  93         if (inode->i_state & I_NEW)
  94                 unlock_new_inode(inode);
  95
  96         return inode;
  97 }
  98
  99 const struct inode_operations ceph_file_iops = {
 100         .permission = ceph_permission,
 101         .setattr = ceph_setattr,
 102         .getattr = ceph_getattr,
 103         .listxattr = ceph_listxattr,
 104         .get_acl = ceph_get_acl,
 105         .set_acl = ceph_set_acl,
 106 };
 107
 108
 109 /*
 110  * We use a 'frag tree' to keep track of the MDS's directory fragments
 111  * for a given inode (usually there is just a single fragment).  We
 112  * need to know when a child frag is delegated to a new MDS, or when
 113  * it is flagged as replicated, so we can direct our requests
 114  * accordingly.
 115  */
 116
 117 /*
 118  * find/create a frag in the tree
 119  */
 120 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 121                                                     u32 f)
 122 {
 123         struct rb_node **p;
 124         struct rb_node *parent = NULL;
 125         struct ceph_inode_frag *frag;
 126         int c;
 127
 128         p = &ci->i_fragtree.rb_node;
 129         while (*p) {
 130                 parent = *p;
 131                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 132                 c = ceph_frag_compare(f, frag->frag);
 133                 if (c < 0)
 134                         p = &(*p)->rb_left;
 135                 else if (c > 0)
 136                         p = &(*p)->rb_right;
 137                 else
 138                         return frag;
 139         }
 140
 141         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 142         if (!frag)
 143                 return ERR_PTR(-ENOMEM);
 144
 145         frag->frag = f;
 146         frag->split_by = 0;
 147         frag->mds = -1;
 148         frag->ndist = 0;
 149
 150         rb_link_node(&frag->node, parent, p);
 151         rb_insert_color(&frag->node, &ci->i_fragtree);
 152
 153         dout("get_or_create_frag added %llx.%llx frag %x\n",
 154              ceph_vinop(&ci->vfs_inode), f);
 155         return frag;
 156 }
 157
 158 /*
 159  * find a specific frag @f
 160  */
 161 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 162 {
 163         struct rb_node *n = ci->i_fragtree.rb_node;
 164
 165         while (n) {
 166                 struct ceph_inode_frag *frag =
 167                         rb_entry(n, struct ceph_inode_frag, node);
 168                 int c = ceph_frag_compare(f, frag->frag);
 169                 if (c < 0)
 170                         n = n->rb_left;
 171                 else if (c > 0)
 172                         n = n->rb_right;
 173                 else
 174                         return frag;
 175         }
 176         return NULL;
 177 }
 178
 179 /*
 180  * Choose frag containing the given value @v.  If @pfrag is
 181  * specified, copy the frag delegation info to the caller if
 182  * it is present.
 183  */
 184 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 185                               struct ceph_inode_frag *pfrag, int *found)
 186 {
 187         u32 t = ceph_frag_make(0, 0);
 188         struct ceph_inode_frag *frag;
 189         unsigned nway, i;
 190         u32 n;
 191
 192         if (found)
 193                 *found = 0;
 194
 195         while (1) {
 196                 WARN_ON(!ceph_frag_contains_value(t, v));
 197                 frag = __ceph_find_frag(ci, t);
 198                 if (!frag)
 199                         break; /* t is a leaf */
 200                 if (frag->split_by == 0) {
 201                         if (pfrag)
 202                                 memcpy(pfrag, frag, sizeof(*pfrag));
 203                         if (found)
 204                                 *found = 1;
 205                         break;
 206                 }
 207
 208                 /* choose child */
 209                 nway = 1 << frag->split_by;
 210                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 211                      frag->split_by, nway);
 212                 for (i = 0; i < nway; i++) {
 213                         n = ceph_frag_make_child(t, frag->split_by, i);
 214                         if (ceph_frag_contains_value(n, v)) {
 215                                 t = n;
 216                                 break;
 217                         }
 218                 }
 219                 BUG_ON(i == nway);
 220         }
 221         dout("choose_frag(%x) = %x\n", v, t);
 222
 223         return t;
 224 }
 225
 226 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 227                      struct ceph_inode_frag *pfrag, int *found)
 228 {
 229         u32 ret;
 230         mutex_lock(&ci->i_fragtree_mutex);
 231         ret = __ceph_choose_frag(ci, v, pfrag, found);
 232         mutex_unlock(&ci->i_fragtree_mutex);
 233         return ret;
 234 }
 235
 236 /*
 237  * Process dirfrag (delegation) info from the mds.  Include leaf
 238  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 239  * branches/splits are included in i_fragtree)
 240  */
 241 static int ceph_fill_dirfrag(struct inode *inode,
 242                              struct ceph_mds_reply_dirfrag *dirinfo)
 243 {
 244         struct ceph_inode_info *ci = ceph_inode(inode);
 245         struct ceph_inode_frag *frag;
 246         u32 id = le32_to_cpu(dirinfo->frag);
 247         int mds = le32_to_cpu(dirinfo->auth);
 248         int ndist = le32_to_cpu(dirinfo->ndist);
 249         int diri_auth = -1;
 250         int i;
 251         int err = 0;
 252
 253         spin_lock(&ci->i_ceph_lock);
 254         if (ci->i_auth_cap)
 255                 diri_auth = ci->i_auth_cap->mds;
 256         spin_unlock(&ci->i_ceph_lock);
 257
 258         if (mds == -1) /* CDIR_AUTH_PARENT */
 259                 mds = diri_auth;
 260
 261         mutex_lock(&ci->i_fragtree_mutex);
 262         if (ndist == 0 && mds == diri_auth) {
 263                 /* no delegation info needed. */
 264                 frag = __ceph_find_frag(ci, id);
 265                 if (!frag)
 266                         goto out;
 267                 if (frag->split_by == 0) {
 268                         /* tree leaf, remove */
 269                         dout("fill_dirfrag removed %llx.%llx frag %x"
 270                              " (no ref)\n", ceph_vinop(inode), id);
 271                         rb_erase(&frag->node, &ci->i_fragtree);
 272                         kfree(frag);
 273                 } else {
 274                         /* tree branch, keep and clear */
 275                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 276                              " referral\n", ceph_vinop(inode), id);
 277                         frag->mds = -1;
 278                         frag->ndist = 0;
 279                 }
 280                 goto out;
 281         }
 282
 283
 284         /* find/add this frag to store mds delegation info */
 285         frag = __get_or_create_frag(ci, id);
 286         if (IS_ERR(frag)) {
 287                 /* this is not the end of the world; we can continue
 288                    with bad/inaccurate delegation info */
 289                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 290                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 291                 err = -ENOMEM;
 292                 goto out;
 293         }
 294
 295         frag->mds = mds;
 296         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 297         for (i = 0; i < frag->ndist; i++)
 298                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 299         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 300              ceph_vinop(inode), frag->frag, frag->ndist);
 301
 302 out:
 303         mutex_unlock(&ci->i_fragtree_mutex);
 304         return err;
 305 }
 306
 307 static int frag_tree_split_cmp(const void *l, const void *r)
 308 {
 309         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 310         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 311         return ceph_frag_compare(le32_to_cpu(ls->frag),
 312                                  le32_to_cpu(rs->frag));
 313 }
 314
 315 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 316 {
 317         if (!frag)
 318                 return f == ceph_frag_make(0, 0);
 319         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 320                 return false;
 321         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 322 }
 323
 324 static int ceph_fill_fragtree(struct inode *inode,
 325                               struct ceph_frag_tree_head *fragtree,
 326                               struct ceph_mds_reply_dirfrag *dirinfo)
 327 {
 328         struct ceph_inode_info *ci = ceph_inode(inode);
 329         struct ceph_inode_frag *frag, *prev_frag = NULL;
 330         struct rb_node *rb_node;
 331         unsigned i, split_by, nsplits;
 332         u32 id;
 333         bool update = false;
 334
 335         mutex_lock(&ci->i_fragtree_mutex);
 336         nsplits = le32_to_cpu(fragtree->nsplits);
 337         if (nsplits != ci->i_fragtree_nsplits) {
 338                 update = true;
 339         } else if (nsplits) {
 340                 i = prandom_u32() % nsplits;
 341                 id = le32_to_cpu(fragtree->splits[i].frag);
 342                 if (!__ceph_find_frag(ci, id))
 343                         update = true;
 344         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 345                 rb_node = rb_first(&ci->i_fragtree);
 346                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 347                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 348                         update = true;
 349         }
 350         if (!update && dirinfo) {
 351                 id = le32_to_cpu(dirinfo->frag);
 352                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 353                         update = true;
 354         }
 355         if (!update)
 356                 goto out_unlock;
 357
 358         if (nsplits > 1) {
 359                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 360                      frag_tree_split_cmp, NULL);
 361         }
 362
 363         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 364         rb_node = rb_first(&ci->i_fragtree);
 365         for (i = 0; i < nsplits; i++) {
 366                 id = le32_to_cpu(fragtree->splits[i].frag);
 367                 split_by = le32_to_cpu(fragtree->splits[i].by);
 368                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 369                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 370                                "frag %x split by %d\n", ceph_vinop(inode),
 371                                i, nsplits, id, split_by);
 372                         continue;
 373                 }
 374                 frag = NULL;
 375                 while (rb_node) {
 376                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 377                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 378                                 if (frag->frag != id)
 379                                         frag = NULL;
 380                                 else
 381                                         rb_node = rb_next(rb_node);
 382                                 break;
 383                         }
 384                         rb_node = rb_next(rb_node);
 385                         /* delete stale split/leaf node */
 386                         if (frag->split_by > 0 ||
 387                             !is_frag_child(frag->frag, prev_frag)) {
 388                                 rb_erase(&frag->node, &ci->i_fragtree);
 389                                 if (frag->split_by > 0)
 390                                         ci->i_fragtree_nsplits--;
 391                                 kfree(frag);
 392                         }
 393                         frag = NULL;
 394                 }
 395                 if (!frag) {
 396                         frag = __get_or_create_frag(ci, id);
 397                         if (IS_ERR(frag))
 398                                 continue;
 399                 }
 400                 if (frag->split_by == 0)
 401                         ci->i_fragtree_nsplits++;
 402                 frag->split_by = split_by;
 403                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 404                 prev_frag = frag;
 405         }
 406         while (rb_node) {
 407                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 408                 rb_node = rb_next(rb_node);
 409                 /* delete stale split/leaf node */
 410                 if (frag->split_by > 0 ||
 411                     !is_frag_child(frag->frag, prev_frag)) {
 412                         rb_erase(&frag->node, &ci->i_fragtree);
 413                         if (frag->split_by > 0)
 414                                 ci->i_fragtree_nsplits--;
 415                         kfree(frag);
 416                 }
 417         }
 418 out_unlock:
 419         mutex_unlock(&ci->i_fragtree_mutex);
 420         return 0;
 421 }
 422
 423 /*
 424  * initialize a newly allocated inode.
 425  */
 426 struct inode *ceph_alloc_inode(struct super_block *sb)
 427 {
 428         struct ceph_inode_info *ci;
 429         int i;
 430
 431         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 432         if (!ci)
 433                 return NULL;
 434
 435         dout("alloc_inode %p\n", &ci->vfs_inode);
 436
 437         spin_lock_init(&ci->i_ceph_lock);
 438
 439         ci->i_version = 0;
 440         ci->i_inline_version = 0;
 441         ci->i_time_warp_seq = 0;
 442         ci->i_ceph_flags = 0;
 443         atomic64_set(&ci->i_ordered_count, 1);
 444         atomic64_set(&ci->i_release_count, 1);
 445         atomic64_set(&ci->i_complete_seq[0], 0);
 446         atomic64_set(&ci->i_complete_seq[1], 0);
 447         ci->i_symlink = NULL;
 448
 449         ci->i_max_bytes = 0;
 450         ci->i_max_files = 0;
 451
 452         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 453         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 454         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 455
 456         ci->i_fragtree = RB_ROOT;
 457         mutex_init(&ci->i_fragtree_mutex);
 458
 459         ci->i_xattrs.blob = NULL;
 460         ci->i_xattrs.prealloc_blob = NULL;
 461         ci->i_xattrs.dirty = false;
 462         ci->i_xattrs.index = RB_ROOT;
 463         ci->i_xattrs.count = 0;
 464         ci->i_xattrs.names_size = 0;
 465         ci->i_xattrs.vals_size = 0;
 466         ci->i_xattrs.version = 0;
 467         ci->i_xattrs.index_version = 0;
 468
 469         ci->i_caps = RB_ROOT;
 470         ci->i_auth_cap = NULL;
 471         ci->i_dirty_caps = 0;
 472         ci->i_flushing_caps = 0;
 473         INIT_LIST_HEAD(&ci->i_dirty_item);
 474         INIT_LIST_HEAD(&ci->i_flushing_item);
 475         ci->i_prealloc_cap_flush = NULL;
 476         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 477         init_waitqueue_head(&ci->i_cap_wq);
 478         ci->i_hold_caps_max = 0;
 479         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 480         INIT_LIST_HEAD(&ci->i_cap_snaps);
 481         ci->i_head_snapc = NULL;
 482         ci->i_snap_caps = 0;
 483
 484         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 485         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 486                 ci->i_nr_by_mode[i] = 0;
 487
 488         mutex_init(&ci->i_truncate_mutex);
 489         ci->i_truncate_seq = 0;
 490         ci->i_truncate_size = 0;
 491         ci->i_truncate_pending = 0;
 492
 493         ci->i_max_size = 0;
 494         ci->i_reported_size = 0;
 495         ci->i_wanted_max_size = 0;
 496         ci->i_requested_max_size = 0;
 497
 498         ci->i_pin_ref = 0;
 499         ci->i_rd_ref = 0;
 500         ci->i_rdcache_ref = 0;
 501         ci->i_wr_ref = 0;
 502         ci->i_wb_ref = 0;
 503         ci->i_fx_ref = 0;
 504         ci->i_wrbuffer_ref = 0;
 505         ci->i_wrbuffer_ref_head = 0;
 506         atomic_set(&ci->i_filelock_ref, 0);
 507         atomic_set(&ci->i_shared_gen, 1);
 508         ci->i_rdcache_gen = 0;
 509         ci->i_rdcache_revoking = 0;
 510
 511         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 512         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 513         spin_lock_init(&ci->i_unsafe_lock);
 514
 515         ci->i_snap_realm = NULL;
 516         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 517         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 518
 519         INIT_WORK(&ci->i_work, ceph_inode_work);
 520         ci->i_work_mask = 0;
 521         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 522
 523         ceph_fscache_inode_init(ci);
 524
 525         ci->i_meta_err = 0;
 526
 527         return &ci->vfs_inode;
 528 }
 529
 530 void ceph_free_inode(struct inode *inode)
 531 {
 532         struct ceph_inode_info *ci = ceph_inode(inode);
 533
 534         kfree(ci->i_symlink);
 535         kmem_cache_free(ceph_inode_cachep, ci);
 536 }
 537
 538 void ceph_evict_inode(struct inode *inode)
 539 {
 540         struct ceph_inode_info *ci = ceph_inode(inode);
 541         struct ceph_inode_frag *frag;
 542         struct rb_node *n;
 543
 544         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 545
 546         truncate_inode_pages_final(&inode->i_data);
 547         clear_inode(inode);
 548
 549         ceph_fscache_unregister_inode_cookie(ci);
 550
 551         __ceph_remove_caps(ci);
 552
 553         if (__ceph_has_any_quota(ci))
 554                 ceph_adjust_quota_realms_count(inode, false);
 555
 556         /*
 557          * we may still have a snap_realm reference if there are stray
 558          * caps in i_snap_caps.
 559          */
 560         if (ci->i_snap_realm) {
 561                 struct ceph_mds_client *mdsc =
 562                                         ceph_inode_to_client(inode)->mdsc;
 563                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 564                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 565                         dout(" dropping residual ref to snap realm %p\n",
 566                              realm);
 567                         spin_lock(&realm->inodes_with_caps_lock);
 568                         list_del_init(&ci->i_snap_realm_item);
 569                         ci->i_snap_realm = NULL;
 570                         if (realm->ino == ci->i_vino.ino)
 571                                 realm->inode = NULL;
 572                         spin_unlock(&realm->inodes_with_caps_lock);
 573                         ceph_put_snap_realm(mdsc, realm);
 574                 } else {
 575                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 576                         ci->i_snap_realm = NULL;
 577                 }
 578         }
 579
 580         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 581                 frag = rb_entry(n, struct ceph_inode_frag, node);
 582                 rb_erase(n, &ci->i_fragtree);
 583                 kfree(frag);
 584         }
 585         ci->i_fragtree_nsplits = 0;
 586
 587         __ceph_destroy_xattrs(ci);
 588         if (ci->i_xattrs.blob)
 589                 ceph_buffer_put(ci->i_xattrs.blob);
 590         if (ci->i_xattrs.prealloc_blob)
 591                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 592
 593         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 594         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 595 }
 596
 597 static inline blkcnt_t calc_inode_blocks(u64 size)
 598 {
 599         return (size + (1<<9) - 1) >> 9;
 600 }
 601
 602 /*
 603  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 604  * careful because either the client or MDS may have more up to date
 605  * info, depending on which capabilities are held, and whether
 606  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 607  * and size are monotonically increasing, except when utimes() or
 608  * truncate() increments the corresponding _seq values.)
 609  */
 610 int ceph_fill_file_size(struct inode *inode, int issued,
 611                         u32 truncate_seq, u64 truncate_size, u64 size)
 612 {
 613         struct ceph_inode_info *ci = ceph_inode(inode);
 614         int queue_trunc = 0;
 615
 616         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 617             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 618                 dout("size %lld -> %llu\n", inode->i_size, size);
 619                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 620                         pr_err("fill_file_size non-zero size for directory\n");
 621                         size = 0;
 622                 }
 623                 i_size_write(inode, size);
 624                 inode->i_blocks = calc_inode_blocks(size);
 625                 ci->i_reported_size = size;
 626                 if (truncate_seq != ci->i_truncate_seq) {
 627                         dout("truncate_seq %u -> %u\n",
 628                              ci->i_truncate_seq, truncate_seq);
 629                         ci->i_truncate_seq = truncate_seq;
 630
 631                         /* the MDS should have revoked these caps */
 632                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 633                                                CEPH_CAP_FILE_RD |
 634                                                CEPH_CAP_FILE_WR |
 635                                                CEPH_CAP_FILE_LAZYIO));
 636                         /*
 637                          * If we hold relevant caps, or in the case where we're
 638                          * not the only client referencing this file and we
 639                          * don't hold those caps, then we need to check whether
 640                          * the file is either opened or mmaped
 641                          */
 642                         if ((issued & (CEPH_CAP_FILE_CACHE|
 643                                        CEPH_CAP_FILE_BUFFER)) ||
 644                             mapping_mapped(inode->i_mapping) ||
 645                             __ceph_is_file_opened(ci)) {
 646                                 ci->i_truncate_pending++;
 647                                 queue_trunc = 1;
 648                         }
 649                 }
 650         }
 651         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 652             ci->i_truncate_size != truncate_size) {
 653                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 654                      truncate_size);
 655                 ci->i_truncate_size = truncate_size;
 656         }
 657
 658         if (queue_trunc)
 659                 ceph_fscache_invalidate(inode);
 660
 661         return queue_trunc;
 662 }
 663
 664 void ceph_fill_file_time(struct inode *inode, int issued,
 665                          u64 time_warp_seq, struct timespec64 *ctime,
 666                          struct timespec64 *mtime, struct timespec64 *atime)
 667 {
 668         struct ceph_inode_info *ci = ceph_inode(inode);
 669         int warn = 0;
 670
 671         if (issued & (CEPH_CAP_FILE_EXCL|
 672                       CEPH_CAP_FILE_WR|
 673                       CEPH_CAP_FILE_BUFFER|
 674                       CEPH_CAP_AUTH_EXCL|
 675                       CEPH_CAP_XATTR_EXCL)) {
 676                 if (ci->i_version == 0 ||
 677                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 678                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 679                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 680                              ctime->tv_sec, ctime->tv_nsec);
 681                         inode->i_ctime = *ctime;
 682                 }
 683                 if (ci->i_version == 0 ||
 684                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 685                         /* the MDS did a utimes() */
 686                         dout("mtime %lld.%09ld -> %lld.%09ld "
 687                              "tw %d -> %d\n",
 688                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 689                              mtime->tv_sec, mtime->tv_nsec,
 690                              ci->i_time_warp_seq, (int)time_warp_seq);
 691
 692                         inode->i_mtime = *mtime;
 693                         inode->i_atime = *atime;
 694                         ci->i_time_warp_seq = time_warp_seq;
 695                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 696                         /* nobody did utimes(); take the max */
 697                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 698                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 699                                      inode->i_mtime.tv_sec,
 700                                      inode->i_mtime.tv_nsec,
 701                                      mtime->tv_sec, mtime->tv_nsec);
 702                                 inode->i_mtime = *mtime;
 703                         }
 704                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 705                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 706                                      inode->i_atime.tv_sec,
 707                                      inode->i_atime.tv_nsec,
 708                                      atime->tv_sec, atime->tv_nsec);
 709                                 inode->i_atime = *atime;
 710                         }
 711                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 712                         /* we did a utimes(); ignore mds values */
 713                 } else {
 714                         warn = 1;
 715                 }
 716         } else {
 717                 /* we have no write|excl caps; whatever the MDS says is true */
 718                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 719                         inode->i_ctime = *ctime;
 720                         inode->i_mtime = *mtime;
 721                         inode->i_atime = *atime;
 722                         ci->i_time_warp_seq = time_warp_seq;
 723                 } else {
 724                         warn = 1;
 725                 }
 726         }
 727         if (warn) /* time_warp_seq shouldn't go backwards */
 728                 dout("%p mds time_warp_seq %llu < %u\n",
 729                      inode, time_warp_seq, ci->i_time_warp_seq);
 730 }
 731
 732 /*
 733  * Populate an inode based on info from mds.  May be called on new or
 734  * existing inodes.
 735  */
 736 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 737                     struct ceph_mds_reply_info_in *iinfo,
 738                     struct ceph_mds_reply_dirfrag *dirinfo,
 739                     struct ceph_mds_session *session, int cap_fmode,
 740                     struct ceph_cap_reservation *caps_reservation)
 741 {
 742         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 743         struct ceph_mds_reply_inode *info = iinfo->in;
 744         struct ceph_inode_info *ci = ceph_inode(inode);
 745         int issued, new_issued, info_caps;
 746         struct timespec64 mtime, atime, ctime;
 747         struct ceph_buffer *xattr_blob = NULL;
 748         struct ceph_buffer *old_blob = NULL;
 749         struct ceph_string *pool_ns = NULL;
 750         struct ceph_cap *new_cap = NULL;
 751         int err = 0;
 752         bool wake = false;
 753         bool queue_trunc = false;
 754         bool new_version = false;
 755         bool fill_inline = false;
 756
 757         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 758              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 759              ci->i_version);
 760
 761         info_caps = le32_to_cpu(info->cap.caps);
 762
 763         /* prealloc new cap struct */
 764         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 765                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 766                 if (!new_cap)
 767                         return -ENOMEM;
 768         }
 769
 770         /*
 771          * prealloc xattr data, if it looks like we'll need it.  only
 772          * if len > 4 (meaning there are actually xattrs; the first 4
 773          * bytes are the xattr count).
 774          */
 775         if (iinfo->xattr_len > 4) {
 776                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 777                 if (!xattr_blob)
 778                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 779                                iinfo->xattr_len);
 780         }
 781
 782         if (iinfo->pool_ns_len > 0)
 783                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 784                                                      iinfo->pool_ns_len);
 785
 786         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 787                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 788
 789         spin_lock(&ci->i_ceph_lock);
 790
 791         /*
 792          * provided version will be odd if inode value is projected,
 793          * even if stable.  skip the update if we have newer stable
 794          * info (ours>=theirs, e.g. due to racing mds replies), unless
 795          * we are getting projected (unstable) info (in which case the
 796          * version is odd, and we want ours>theirs).
 797          *   us   them
 798          *   2    2     skip
 799          *   3    2     skip
 800          *   3    3     update
 801          */
 802         if (ci->i_version == 0 ||
 803             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 804              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 805                 new_version = true;
 806
 807         /* Update change_attribute */
 808         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 809
 810         __ceph_caps_issued(ci, &issued);
 811         issued |= __ceph_caps_dirty(ci);
 812         new_issued = ~issued & info_caps;
 813
 814         /* update inode */
 815         inode->i_rdev = le32_to_cpu(info->rdev);
 816         /* directories have fl_stripe_unit set to zero */
 817         if (le32_to_cpu(info->layout.fl_stripe_unit))
 818                 inode->i_blkbits =
 819                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 820         else
 821                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 822
 823         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 824
 825         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 826             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 827                 inode->i_mode = le32_to_cpu(info->mode);
 828                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 829                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 830                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 831                      from_kuid(&init_user_ns, inode->i_uid),
 832                      from_kgid(&init_user_ns, inode->i_gid));
 833                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 834                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 835         }
 836
 837         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 838             (issued & CEPH_CAP_LINK_EXCL) == 0)
 839                 set_nlink(inode, le32_to_cpu(info->nlink));
 840
 841         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 842                 /* be careful with mtime, atime, size */
 843                 ceph_decode_timespec64(&atime, &info->atime);
 844                 ceph_decode_timespec64(&mtime, &info->mtime);
 845                 ceph_decode_timespec64(&ctime, &info->ctime);
 846                 ceph_fill_file_time(inode, issued,
 847                                 le32_to_cpu(info->time_warp_seq),
 848                                 &ctime, &mtime, &atime);
 849         }
 850
 851         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 852                 ci->i_files = le64_to_cpu(info->files);
 853                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 854         }
 855
 856         if (new_version ||
 857             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 858                 s64 old_pool = ci->i_layout.pool_id;
 859                 struct ceph_string *old_ns;
 860
 861                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 862                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 863                                         lockdep_is_held(&ci->i_ceph_lock));
 864                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 865
 866                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 867                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 868
 869                 pool_ns = old_ns;
 870
 871                 queue_trunc = ceph_fill_file_size(inode, issued,
 872                                         le32_to_cpu(info->truncate_seq),
 873                                         le64_to_cpu(info->truncate_size),
 874                                         le64_to_cpu(info->size));
 875                 /* only update max_size on auth cap */
 876                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 877                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 878                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 879                                         le64_to_cpu(info->max_size));
 880                         ci->i_max_size = le64_to_cpu(info->max_size);
 881                 }
 882         }
 883
 884         /* layout and rstat are not tracked by capability, update them if
 885          * the inode info is from auth mds */
 886         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 887                 if (S_ISDIR(inode->i_mode)) {
 888                         ci->i_dir_layout = iinfo->dir_layout;
 889                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 890                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 891                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 892                         ci->i_dir_pin = iinfo->dir_pin;
 893                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 894                 }
 895         }
 896
 897         /* xattrs */
 898         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 899         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 900             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 901                 if (ci->i_xattrs.blob)
 902                         old_blob = ci->i_xattrs.blob;
 903                 ci->i_xattrs.blob = xattr_blob;
 904                 if (xattr_blob)
 905                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 906                                iinfo->xattr_data, iinfo->xattr_len);
 907                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 908                 ceph_forget_all_cached_acls(inode);
 909                 ceph_security_invalidate_secctx(inode);
 910                 xattr_blob = NULL;
 911         }
 912
 913         /* finally update i_version */
 914         if (le64_to_cpu(info->version) > ci->i_version)
 915                 ci->i_version = le64_to_cpu(info->version);
 916
 917         inode->i_mapping->a_ops = &ceph_aops;
 918
 919         switch (inode->i_mode & S_IFMT) {
 920         case S_IFIFO:
 921         case S_IFBLK:
 922         case S_IFCHR:
 923         case S_IFSOCK:
 924                 inode->i_blkbits = PAGE_SHIFT;
 925                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 926                 inode->i_op = &ceph_file_iops;
 927                 break;
 928         case S_IFREG:
 929                 inode->i_op = &ceph_file_iops;
 930                 inode->i_fop = &ceph_file_fops;
 931                 break;
 932         case S_IFLNK:
 933                 inode->i_op = &ceph_symlink_iops;
 934                 if (!ci->i_symlink) {
 935                         u32 symlen = iinfo->symlink_len;
 936                         char *sym;
 937
 938                         spin_unlock(&ci->i_ceph_lock);
 939
 940                         if (symlen != i_size_read(inode)) {
 941                                 pr_err("%s %llx.%llx BAD symlink "
 942                                         "size %lld\n", __func__,
 943                                         ceph_vinop(inode),
 944                                         i_size_read(inode));
 945                                 i_size_write(inode, symlen);
 946                                 inode->i_blocks = calc_inode_blocks(symlen);
 947                         }
 948
 949                         err = -ENOMEM;
 950                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 951                         if (!sym)
 952                                 goto out;
 953
 954                         spin_lock(&ci->i_ceph_lock);
 955                         if (!ci->i_symlink)
 956                                 ci->i_symlink = sym;
 957                         else
 958                                 kfree(sym); /* lost a race */
 959                 }
 960                 inode->i_link = ci->i_symlink;
 961                 break;
 962         case S_IFDIR:
 963                 inode->i_op = &ceph_dir_iops;
 964                 inode->i_fop = &ceph_dir_fops;
 965                 break;
 966         default:
 967                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
 968                        ceph_vinop(inode), inode->i_mode);
 969         }
 970
 971         /* were we issued a capability? */
 972         if (info_caps) {
 973                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 974                         ceph_add_cap(inode, session,
 975                                      le64_to_cpu(info->cap.cap_id),
 976                                      info_caps,
 977                                      le32_to_cpu(info->cap.wanted),
 978                                      le32_to_cpu(info->cap.seq),
 979                                      le32_to_cpu(info->cap.mseq),
 980                                      le64_to_cpu(info->cap.realm),
 981                                      info->cap.flags, &new_cap);
 982
 983                         /* set dir completion flag? */
 984                         if (S_ISDIR(inode->i_mode) &&
 985                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 986                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 987                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 988                             !__ceph_dir_is_complete(ci)) {
 989                                 dout(" marking %p complete (empty)\n", inode);
 990                                 i_size_write(inode, 0);
 991                                 __ceph_dir_set_complete(ci,
 992                                         atomic64_read(&ci->i_release_count),
 993                                         atomic64_read(&ci->i_ordered_count));
 994                         }
 995
 996                         wake = true;
 997                 } else {
 998                         dout(" %p got snap_caps %s\n", inode,
 999                              ceph_cap_string(info_caps));
1000                         ci->i_snap_caps |= info_caps;
1001                 }
1002         }
1003
1004         if (iinfo->inline_version > 0 &&
1005             iinfo->inline_version >= ci->i_inline_version) {
1006                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1007                 ci->i_inline_version = iinfo->inline_version;
1008                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1009                     (locked_page || (info_caps & cache_caps)))
1010                         fill_inline = true;
1011         }
1012
1013         if (cap_fmode >= 0) {
1014                 if (!info_caps)
1015                         pr_warn("mds issued no caps on %llx.%llx\n",
1016                                 ceph_vinop(inode));
1017                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1018         }
1019
1020         spin_unlock(&ci->i_ceph_lock);
1021
1022         if (fill_inline)
1023                 ceph_fill_inline_data(inode, locked_page,
1024                                       iinfo->inline_data, iinfo->inline_len);
1025
1026         if (wake)
1027                 wake_up_all(&ci->i_cap_wq);
1028
1029         /* queue truncate if we saw i_size decrease */
1030         if (queue_trunc)
1031                 ceph_queue_vmtruncate(inode);
1032
1033         /* populate frag tree */
1034         if (S_ISDIR(inode->i_mode))
1035                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1036
1037         /* update delegation info? */
1038         if (dirinfo)
1039                 ceph_fill_dirfrag(inode, dirinfo);
1040
1041         err = 0;
1042 out:
1043         if (new_cap)
1044                 ceph_put_cap(mdsc, new_cap);
1045         ceph_buffer_put(old_blob);
1046         ceph_buffer_put(xattr_blob);
1047         ceph_put_string(pool_ns);
1048         return err;
1049 }
1050
1051 /*
1052  * caller should hold session s_mutex and dentry->d_lock.
1053  */
1054 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1055                                   struct ceph_mds_reply_lease *lease,
1056                                   struct ceph_mds_session *session,
1057                                   unsigned long from_time,
1058                                   struct ceph_mds_session **old_lease_session)
1059 {
1060         struct ceph_dentry_info *di = ceph_dentry(dentry);
1061         unsigned mask = le16_to_cpu(lease->mask);
1062         long unsigned duration = le32_to_cpu(lease->duration_ms);
1063         long unsigned ttl = from_time + (duration * HZ) / 1000;
1064         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1065
1066         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1067              dentry, duration, ttl);
1068
1069         /* only track leases on regular dentries */
1070         if (ceph_snap(dir) != CEPH_NOSNAP)
1071                 return;
1072
1073         if (mask & CEPH_LEASE_PRIMARY_LINK)
1074                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1075         else
1076                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1077
1078         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1079         if (!(mask & CEPH_LEASE_VALID)) {
1080                 __ceph_dentry_dir_lease_touch(di);
1081                 return;
1082         }
1083
1084         if (di->lease_gen == session->s_cap_gen &&
1085             time_before(ttl, di->time))
1086                 return;  /* we already have a newer lease. */
1087
1088         if (di->lease_session && di->lease_session != session) {
1089                 *old_lease_session = di->lease_session;
1090                 di->lease_session = NULL;
1091         }
1092
1093         if (!di->lease_session)
1094                 di->lease_session = ceph_get_mds_session(session);
1095         di->lease_gen = session->s_cap_gen;
1096         di->lease_seq = le32_to_cpu(lease->seq);
1097         di->lease_renew_after = half_ttl;
1098         di->lease_renew_from = 0;
1099         di->time = ttl;
1100
1101         __ceph_dentry_lease_touch(di);
1102 }
1103
1104 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1105                                         struct ceph_mds_reply_lease *lease,
1106                                         struct ceph_mds_session *session,
1107                                         unsigned long from_time)
1108 {
1109         struct ceph_mds_session *old_lease_session = NULL;
1110         spin_lock(&dentry->d_lock);
1111         __update_dentry_lease(dir, dentry, lease, session, from_time,
1112                               &old_lease_session);
1113         spin_unlock(&dentry->d_lock);
1114         if (old_lease_session)
1115                 ceph_put_mds_session(old_lease_session);
1116 }
1117
1118 /*
1119  * update dentry lease without having parent inode locked
1120  */
1121 static void update_dentry_lease_careful(struct dentry *dentry,
1122                                         struct ceph_mds_reply_lease *lease,
1123                                         struct ceph_mds_session *session,
1124                                         unsigned long from_time,
1125                                         char *dname, u32 dname_len,
1126                                         struct ceph_vino *pdvino,
1127                                         struct ceph_vino *ptvino)
1128
1129 {
1130         struct inode *dir;
1131         struct ceph_mds_session *old_lease_session = NULL;
1132
1133         spin_lock(&dentry->d_lock);
1134         /* make sure dentry's name matches target */
1135         if (dentry->d_name.len != dname_len ||
1136             memcmp(dentry->d_name.name, dname, dname_len))
1137                 goto out_unlock;
1138
1139         dir = d_inode(dentry->d_parent);
1140         /* make sure parent matches dvino */
1141         if (!ceph_ino_compare(dir, pdvino))
1142                 goto out_unlock;
1143
1144         /* make sure dentry's inode matches target. NULL ptvino means that
1145          * we expect a negative dentry */
1146         if (ptvino) {
1147                 if (d_really_is_negative(dentry))
1148                         goto out_unlock;
1149                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1150                         goto out_unlock;
1151         } else {
1152                 if (d_really_is_positive(dentry))
1153                         goto out_unlock;
1154         }
1155
1156         __update_dentry_lease(dir, dentry, lease, session,
1157                               from_time, &old_lease_session);
1158 out_unlock:
1159         spin_unlock(&dentry->d_lock);
1160         if (old_lease_session)
1161                 ceph_put_mds_session(old_lease_session);
1162 }
1163
1164 /*
1165  * splice a dentry to an inode.
1166  * caller must hold directory i_mutex for this to be safe.
1167  */
1168 static int splice_dentry(struct dentry **pdn, struct inode *in)
1169 {
1170         struct dentry *dn = *pdn;
1171         struct dentry *realdn;
1172
1173         BUG_ON(d_inode(dn));
1174
1175         if (S_ISDIR(in->i_mode)) {
1176                 /* If inode is directory, d_splice_alias() below will remove
1177                  * 'realdn' from its origin parent. We need to ensure that
1178                  * origin parent's readdir cache will not reference 'realdn'
1179                  */
1180                 realdn = d_find_any_alias(in);
1181                 if (realdn) {
1182                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1183                         spin_lock(&realdn->d_lock);
1184
1185                         realdn->d_op->d_prune(realdn);
1186
1187                         di->time = jiffies;
1188                         di->lease_shared_gen = 0;
1189                         di->offset = 0;
1190
1191                         spin_unlock(&realdn->d_lock);
1192                         dput(realdn);
1193                 }
1194         }
1195
1196         /* dn must be unhashed */
1197         if (!d_unhashed(dn))
1198                 d_drop(dn);
1199         realdn = d_splice_alias(in, dn);
1200         if (IS_ERR(realdn)) {
1201                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1202                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1203                 return PTR_ERR(realdn);
1204         }
1205
1206         if (realdn) {
1207                 dout("dn %p (%d) spliced with %p (%d) "
1208                      "inode %p ino %llx.%llx\n",
1209                      dn, d_count(dn),
1210                      realdn, d_count(realdn),
1211                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1212                 dput(dn);
1213                 *pdn = realdn;
1214         } else {
1215                 BUG_ON(!ceph_dentry(dn));
1216                 dout("dn %p attached to %p ino %llx.%llx\n",
1217                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1218         }
1219         return 0;
1220 }
1221
1222 /*
1223  * Incorporate results into the local cache.  This is either just
1224  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1225  * after a lookup).
1226  *
1227  * A reply may contain
1228  *         a directory inode along with a dentry.
1229  *  and/or a target inode
1230  *
1231  * Called with snap_rwsem (read).
1232  */
1233 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1234 {
1235         struct ceph_mds_session *session = req->r_session;
1236         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1237         struct inode *in = NULL;
1238         struct ceph_vino tvino, dvino;
1239         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1240         int err = 0;
1241
1242         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1243              rinfo->head->is_dentry, rinfo->head->is_target);
1244
1245         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1246                 dout("fill_trace reply is empty!\n");
1247                 if (rinfo->head->result == 0 && req->r_parent)
1248                         ceph_invalidate_dir_request(req);
1249                 return 0;
1250         }
1251
1252         if (rinfo->head->is_dentry) {
1253                 struct inode *dir = req->r_parent;
1254
1255                 if (dir) {
1256                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1257                                               rinfo->dirfrag, session, -1,
1258                                               &req->r_caps_reservation);
1259                         if (err < 0)
1260                                 goto done;
1261                 } else {
1262                         WARN_ON_ONCE(1);
1263                 }
1264
1265                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1266                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1267                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1268                         struct qstr dname;
1269                         struct dentry *dn, *parent;
1270
1271                         BUG_ON(!rinfo->head->is_target);
1272                         BUG_ON(req->r_dentry);
1273
1274                         parent = d_find_any_alias(dir);
1275                         BUG_ON(!parent);
1276
1277                         dname.name = rinfo->dname;
1278                         dname.len = rinfo->dname_len;
1279                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1280                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1281                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1282 retry_lookup:
1283                         dn = d_lookup(parent, &dname);
1284                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1285                              parent, dname.len, dname.name, dn);
1286
1287                         if (!dn) {
1288                                 dn = d_alloc(parent, &dname);
1289                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1290                                      dname.len, dname.name, dn);
1291                                 if (!dn) {
1292                                         dput(parent);
1293                                         err = -ENOMEM;
1294                                         goto done;
1295                                 }
1296                                 err = 0;
1297                         } else if (d_really_is_positive(dn) &&
1298                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1299                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1300                                 dout(" dn %p points to wrong inode %p\n",
1301                                      dn, d_inode(dn));
1302                                 ceph_dir_clear_ordered(dir);
1303                                 d_delete(dn);
1304                                 dput(dn);
1305                                 goto retry_lookup;
1306                         }
1307
1308                         req->r_dentry = dn;
1309                         dput(parent);
1310                 }
1311         }
1312
1313         if (rinfo->head->is_target) {
1314                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1315                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1316
1317                 in = ceph_get_inode(sb, tvino);
1318                 if (IS_ERR(in)) {
1319                         err = PTR_ERR(in);
1320                         goto done;
1321                 }
1322
1323                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1324                                 NULL, session,
1325                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1326                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1327                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1328                                 &req->r_caps_reservation);
1329                 if (err < 0) {
1330                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1331                                 in, ceph_vinop(in));
1332                         if (in->i_state & I_NEW)
1333                                 discard_new_inode(in);
1334                         goto done;
1335                 }
1336                 req->r_target_inode = in;
1337                 if (in->i_state & I_NEW)
1338                         unlock_new_inode(in);
1339         }
1340
1341         /*
1342          * ignore null lease/binding on snapdir ENOENT, or else we
1343          * will have trouble splicing in the virtual snapdir later
1344          */
1345         if (rinfo->head->is_dentry &&
1346             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1347             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1348             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1349                                                fsc->mount_options->snapdir_name,
1350                                                req->r_dentry->d_name.len))) {
1351                 /*
1352                  * lookup link rename   : null -> possibly existing inode
1353                  * mknod symlink mkdir  : null -> new inode
1354                  * unlink               : linked -> null
1355                  */
1356                 struct inode *dir = req->r_parent;
1357                 struct dentry *dn = req->r_dentry;
1358                 bool have_dir_cap, have_lease;
1359
1360                 BUG_ON(!dn);
1361                 BUG_ON(!dir);
1362                 BUG_ON(d_inode(dn->d_parent) != dir);
1363
1364                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1365                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1366
1367                 BUG_ON(ceph_ino(dir) != dvino.ino);
1368                 BUG_ON(ceph_snap(dir) != dvino.snap);
1369
1370                 /* do we have a lease on the whole dir? */
1371                 have_dir_cap =
1372                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1373                          CEPH_CAP_FILE_SHARED);
1374
1375                 /* do we have a dn lease? */
1376                 have_lease = have_dir_cap ||
1377                         le32_to_cpu(rinfo->dlease->duration_ms);
1378                 if (!have_lease)
1379                         dout("fill_trace  no dentry lease or dir cap\n");
1380
1381                 /* rename? */
1382                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1383                         struct inode *olddir = req->r_old_dentry_dir;
1384                         BUG_ON(!olddir);
1385
1386                         dout(" src %p '%pd' dst %p '%pd'\n",
1387                              req->r_old_dentry,
1388                              req->r_old_dentry,
1389                              dn, dn);
1390                         dout("fill_trace doing d_move %p -> %p\n",
1391                              req->r_old_dentry, dn);
1392
1393                         /* d_move screws up sibling dentries' offsets */
1394                         ceph_dir_clear_ordered(dir);
1395                         ceph_dir_clear_ordered(olddir);
1396
1397                         d_move(req->r_old_dentry, dn);
1398                         dout(" src %p '%pd' dst %p '%pd'\n",
1399                              req->r_old_dentry,
1400                              req->r_old_dentry,
1401                              dn, dn);
1402
1403                         /* ensure target dentry is invalidated, despite
1404                            rehashing bug in vfs_rename_dir */
1405                         ceph_invalidate_dentry_lease(dn);
1406
1407                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1408                              ceph_dentry(req->r_old_dentry)->offset);
1409
1410                         /* swap r_dentry and r_old_dentry in case that
1411                          * splice_dentry() gets called later. This is safe
1412                          * because no other place will use them */
1413                         req->r_dentry = req->r_old_dentry;
1414                         req->r_old_dentry = dn;
1415                         dn = req->r_dentry;
1416                 }
1417
1418                 /* null dentry? */
1419                 if (!rinfo->head->is_target) {
1420                         dout("fill_trace null dentry\n");
1421                         if (d_really_is_positive(dn)) {
1422                                 dout("d_delete %p\n", dn);
1423                                 ceph_dir_clear_ordered(dir);
1424                                 d_delete(dn);
1425                         } else if (have_lease) {
1426                                 if (d_unhashed(dn))
1427                                         d_add(dn, NULL);
1428                                 update_dentry_lease(dir, dn,
1429                                                     rinfo->dlease, session,
1430                                                     req->r_request_started);
1431                         }
1432                         goto done;
1433                 }
1434
1435                 /* attach proper inode */
1436                 if (d_really_is_negative(dn)) {
1437                         ceph_dir_clear_ordered(dir);
1438                         ihold(in);
1439                         err = splice_dentry(&req->r_dentry, in);
1440                         if (err < 0)
1441                                 goto done;
1442                         dn = req->r_dentry;  /* may have spliced */
1443                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1444                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1445                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1446                              ceph_vinop(in));
1447                         d_invalidate(dn);
1448                         have_lease = false;
1449                 }
1450
1451                 if (have_lease) {
1452                         update_dentry_lease(dir, dn,
1453                                             rinfo->dlease, session,
1454                                             req->r_request_started);
1455                 }
1456                 dout(" final dn %p\n", dn);
1457         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1458                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1459                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1460                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1461                 struct inode *dir = req->r_parent;
1462
1463                 /* fill out a snapdir LOOKUPSNAP dentry */
1464                 BUG_ON(!dir);
1465                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1466                 BUG_ON(!req->r_dentry);
1467                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1468                 ceph_dir_clear_ordered(dir);
1469                 ihold(in);
1470                 err = splice_dentry(&req->r_dentry, in);
1471                 if (err < 0)
1472                         goto done;
1473         } else if (rinfo->head->is_dentry && req->r_dentry) {
1474                 /* parent inode is not locked, be carefull */
1475                 struct ceph_vino *ptvino = NULL;
1476                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1477                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1478                 if (rinfo->head->is_target) {
1479                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1480                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1481                         ptvino = &tvino;
1482                 }
1483                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1484                                             session, req->r_request_started,
1485                                             rinfo->dname, rinfo->dname_len,
1486                                             &dvino, ptvino);
1487         }
1488 done:
1489         dout("fill_trace done err=%d\n", err);
1490         return err;
1491 }
1492
1493 /*
1494  * Prepopulate our cache with readdir results, leases, etc.
1495  */
1496 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1497                                            struct ceph_mds_session *session)
1498 {
1499         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1500         int i, err = 0;
1501
1502         for (i = 0; i < rinfo->dir_nr; i++) {
1503                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1504                 struct ceph_vino vino;
1505                 struct inode *in;
1506                 int rc;
1507
1508                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1509                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1510
1511                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1512                 if (IS_ERR(in)) {
1513                         err = PTR_ERR(in);
1514                         dout("new_inode badness got %d\n", err);
1515                         continue;
1516                 }
1517                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1518                                      -1, &req->r_caps_reservation);
1519                 if (rc < 0) {
1520                         pr_err("ceph_fill_inode badness on %p got %d\n",
1521                                in, rc);
1522                         err = rc;
1523                         if (in->i_state & I_NEW) {
1524                                 ihold(in);
1525                                 discard_new_inode(in);
1526                         }
1527                 } else if (in->i_state & I_NEW) {
1528                         unlock_new_inode(in);
1529                 }
1530
1531                 /* avoid calling iput_final() in mds dispatch threads */
1532                 ceph_async_iput(in);
1533         }
1534
1535         return err;
1536 }
1537
1538 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1539 {
1540         if (ctl->page) {
1541                 kunmap(ctl->page);
1542                 put_page(ctl->page);
1543                 ctl->page = NULL;
1544         }
1545 }
1546
1547 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1548                               struct ceph_readdir_cache_control *ctl,
1549                               struct ceph_mds_request *req)
1550 {
1551         struct ceph_inode_info *ci = ceph_inode(dir);
1552         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1553         unsigned idx = ctl->index % nsize;
1554         pgoff_t pgoff = ctl->index / nsize;
1555
1556         if (!ctl->page || pgoff != page_index(ctl->page)) {
1557                 ceph_readdir_cache_release(ctl);
1558                 if (idx == 0)
1559                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1560                 else
1561                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1562                 if (!ctl->page) {
1563                         ctl->index = -1;
1564                         return idx == 0 ? -ENOMEM : 0;
1565                 }
1566                 /* reading/filling the cache are serialized by
1567                  * i_mutex, no need to use page lock */
1568                 unlock_page(ctl->page);
1569                 ctl->dentries = kmap(ctl->page);
1570                 if (idx == 0)
1571                         memset(ctl->dentries, 0, PAGE_SIZE);
1572         }
1573
1574         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1575             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1576                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1577                 ctl->dentries[idx] = dn;
1578                 ctl->index++;
1579         } else {
1580                 dout("disable readdir cache\n");
1581                 ctl->index = -1;
1582         }
1583         return 0;
1584 }
1585
1586 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1587                              struct ceph_mds_session *session)
1588 {
1589         struct dentry *parent = req->r_dentry;
1590         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1591         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1592         struct qstr dname;
1593         struct dentry *dn;
1594         struct inode *in;
1595         int err = 0, skipped = 0, ret, i;
1596         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1597         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1598         u32 last_hash = 0;
1599         u32 fpos_offset;
1600         struct ceph_readdir_cache_control cache_ctl = {};
1601
1602         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1603                 return readdir_prepopulate_inodes_only(req, session);
1604
1605         if (rinfo->hash_order) {
1606                 if (req->r_path2) {
1607                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1608                                                   req->r_path2,
1609                                                   strlen(req->r_path2));
1610                         last_hash = ceph_frag_value(last_hash);
1611                 } else if (rinfo->offset_hash) {
1612                         /* mds understands offset_hash */
1613                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1614                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1615                 }
1616         }
1617
1618         if (rinfo->dir_dir &&
1619             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1620                 dout("readdir_prepopulate got new frag %x -> %x\n",
1621                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1622                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1623                 if (!rinfo->hash_order)
1624                         req->r_readdir_offset = 2;
1625         }
1626
1627         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1628                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1629                      rinfo->dir_nr, parent);
1630         } else {
1631                 dout("readdir_prepopulate %d items under dn %p\n",
1632                      rinfo->dir_nr, parent);
1633                 if (rinfo->dir_dir)
1634                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1635
1636                 if (ceph_frag_is_leftmost(frag) &&
1637                     req->r_readdir_offset == 2 &&
1638                     !(rinfo->hash_order && last_hash)) {
1639                         /* note dir version at start of readdir so we can
1640                          * tell if any dentries get dropped */
1641                         req->r_dir_release_cnt =
1642                                 atomic64_read(&ci->i_release_count);
1643                         req->r_dir_ordered_cnt =
1644                                 atomic64_read(&ci->i_ordered_count);
1645                         req->r_readdir_cache_idx = 0;
1646                 }
1647         }
1648
1649         cache_ctl.index = req->r_readdir_cache_idx;
1650         fpos_offset = req->r_readdir_offset;
1651
1652         /* FIXME: release caps/leases if error occurs */
1653         for (i = 0; i < rinfo->dir_nr; i++) {
1654                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1655                 struct ceph_vino tvino;
1656
1657                 dname.name = rde->name;
1658                 dname.len = rde->name_len;
1659                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1660
1661                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1662                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1663
1664                 if (rinfo->hash_order) {
1665                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1666                                                  rde->name, rde->name_len);
1667                         hash = ceph_frag_value(hash);
1668                         if (hash != last_hash)
1669                                 fpos_offset = 2;
1670                         last_hash = hash;
1671                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1672                 } else {
1673                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1674                 }
1675
1676 retry_lookup:
1677                 dn = d_lookup(parent, &dname);
1678                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1679                      parent, dname.len, dname.name, dn);
1680
1681                 if (!dn) {
1682                         dn = d_alloc(parent, &dname);
1683                         dout("d_alloc %p '%.*s' = %p\n", parent,
1684                              dname.len, dname.name, dn);
1685                         if (!dn) {
1686                                 dout("d_alloc badness\n");
1687                                 err = -ENOMEM;
1688                                 goto out;
1689                         }
1690                 } else if (d_really_is_positive(dn) &&
1691                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1692                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1693                         struct ceph_dentry_info *di = ceph_dentry(dn);
1694                         dout(" dn %p points to wrong inode %p\n",
1695                              dn, d_inode(dn));
1696
1697                         spin_lock(&dn->d_lock);
1698                         if (di->offset > 0 &&
1699                             di->lease_shared_gen ==
1700                             atomic_read(&ci->i_shared_gen)) {
1701                                 __ceph_dir_clear_ordered(ci);
1702                                 di->offset = 0;
1703                         }
1704                         spin_unlock(&dn->d_lock);
1705
1706                         d_delete(dn);
1707                         dput(dn);
1708                         goto retry_lookup;
1709                 }
1710
1711                 /* inode */
1712                 if (d_really_is_positive(dn)) {
1713                         in = d_inode(dn);
1714                 } else {
1715                         in = ceph_get_inode(parent->d_sb, tvino);
1716                         if (IS_ERR(in)) {
1717                                 dout("new_inode badness\n");
1718                                 d_drop(dn);
1719                                 dput(dn);
1720                                 err = PTR_ERR(in);
1721                                 goto out;
1722                         }
1723                 }
1724
1725                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1726                                       -1, &req->r_caps_reservation);
1727                 if (ret < 0) {
1728                         pr_err("ceph_fill_inode badness on %p\n", in);
1729                         if (d_really_is_negative(dn)) {
1730                                 /* avoid calling iput_final() in mds
1731                                  * dispatch threads */
1732                                 if (in->i_state & I_NEW) {
1733                                         ihold(in);
1734                                         discard_new_inode(in);
1735                                 }
1736                                 ceph_async_iput(in);
1737                         }
1738                         d_drop(dn);
1739                         err = ret;
1740                         goto next_item;
1741                 }
1742                 if (in->i_state & I_NEW)
1743                         unlock_new_inode(in);
1744
1745                 if (d_really_is_negative(dn)) {
1746                         if (ceph_security_xattr_deadlock(in)) {
1747                                 dout(" skip splicing dn %p to inode %p"
1748                                      " (security xattr deadlock)\n", dn, in);
1749                                 ceph_async_iput(in);
1750                                 skipped++;
1751                                 goto next_item;
1752                         }
1753
1754                         err = splice_dentry(&dn, in);
1755                         if (err < 0)
1756                                 goto next_item;
1757                 }
1758
1759                 ceph_dentry(dn)->offset = rde->offset;
1760
1761                 update_dentry_lease(d_inode(parent), dn,
1762                                     rde->lease, req->r_session,
1763                                     req->r_request_started);
1764
1765                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1766                         ret = fill_readdir_cache(d_inode(parent), dn,
1767                                                  &cache_ctl, req);
1768                         if (ret < 0)
1769                                 err = ret;
1770                 }
1771 next_item:
1772                 dput(dn);
1773         }
1774 out:
1775         if (err == 0 && skipped == 0) {
1776                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1777                 req->r_readdir_cache_idx = cache_ctl.index;
1778         }
1779         ceph_readdir_cache_release(&cache_ctl);
1780         dout("readdir_prepopulate done\n");
1781         return err;
1782 }
1783
1784 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1785 {
1786         struct ceph_inode_info *ci = ceph_inode(inode);
1787         bool ret;
1788
1789         spin_lock(&ci->i_ceph_lock);
1790         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1791         i_size_write(inode, size);
1792         inode->i_blocks = calc_inode_blocks(size);
1793
1794         ret = __ceph_should_report_size(ci);
1795
1796         spin_unlock(&ci->i_ceph_lock);
1797         return ret;
1798 }
1799
1800 /*
1801  * Put reference to inode, but avoid calling iput_final() in current thread.
1802  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1803  * some contexts.
1804  */
1805 void ceph_async_iput(struct inode *inode)
1806 {
1807         if (!inode)
1808                 return;
1809         for (;;) {
1810                 if (atomic_add_unless(&inode->i_count, -1, 1))
1811                         break;
1812                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1813                                &ceph_inode(inode)->i_work))
1814                         break;
1815                 /* queue work failed, i_count must be at least 2 */
1816         }
1817 }
1818
1819 /*
1820  * Write back inode data in a worker thread.  (This can't be done
1821  * in the message handler context.)
1822  */
1823 void ceph_queue_writeback(struct inode *inode)
1824 {
1825         struct ceph_inode_info *ci = ceph_inode(inode);
1826         set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1827
1828         ihold(inode);
1829         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1830                        &ci->i_work)) {
1831                 dout("ceph_queue_writeback %p\n", inode);
1832         } else {
1833                 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1834                      inode, ci->i_work_mask);
1835                 iput(inode);
1836         }
1837 }
1838
1839 /*
1840  * queue an async invalidation
1841  */
1842 void ceph_queue_invalidate(struct inode *inode)
1843 {
1844         struct ceph_inode_info *ci = ceph_inode(inode);
1845         set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1846
1847         ihold(inode);
1848         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1849                        &ceph_inode(inode)->i_work)) {
1850                 dout("ceph_queue_invalidate %p\n", inode);
1851         } else {
1852                 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1853                      inode, ci->i_work_mask);
1854                 iput(inode);
1855         }
1856 }
1857
1858 /*
1859  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1860  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1861  */
1862 void ceph_queue_vmtruncate(struct inode *inode)
1863 {
1864         struct ceph_inode_info *ci = ceph_inode(inode);
1865         set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1866
1867         ihold(inode);
1868         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1869                        &ci->i_work)) {
1870                 dout("ceph_queue_vmtruncate %p\n", inode);
1871         } else {
1872                 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1873                      inode, ci->i_work_mask);
1874                 iput(inode);
1875         }
1876 }
1877
1878 static void ceph_do_invalidate_pages(struct inode *inode)
1879 {
1880         struct ceph_inode_info *ci = ceph_inode(inode);
1881         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1882         u32 orig_gen;
1883         int check = 0;
1884
1885         mutex_lock(&ci->i_truncate_mutex);
1886
1887         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1888                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1889                                     inode, ceph_ino(inode));
1890                 mapping_set_error(inode->i_mapping, -EIO);
1891                 truncate_pagecache(inode, 0);
1892                 mutex_unlock(&ci->i_truncate_mutex);
1893                 goto out;
1894         }
1895
1896         spin_lock(&ci->i_ceph_lock);
1897         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1898              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1899         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1900                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1901                         check = 1;
1902                 spin_unlock(&ci->i_ceph_lock);
1903                 mutex_unlock(&ci->i_truncate_mutex);
1904                 goto out;
1905         }
1906         orig_gen = ci->i_rdcache_gen;
1907         spin_unlock(&ci->i_ceph_lock);
1908
1909         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1910                 pr_err("invalidate_pages %p fails\n", inode);
1911         }
1912
1913         spin_lock(&ci->i_ceph_lock);
1914         if (orig_gen == ci->i_rdcache_gen &&
1915             orig_gen == ci->i_rdcache_revoking) {
1916                 dout("invalidate_pages %p gen %d successful\n", inode,
1917                      ci->i_rdcache_gen);
1918                 ci->i_rdcache_revoking--;
1919                 check = 1;
1920         } else {
1921                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1922                      inode, orig_gen, ci->i_rdcache_gen,
1923                      ci->i_rdcache_revoking);
1924                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1925                         check = 1;
1926         }
1927         spin_unlock(&ci->i_ceph_lock);
1928         mutex_unlock(&ci->i_truncate_mutex);
1929 out:
1930         if (check)
1931                 ceph_check_caps(ci, 0, NULL);
1932 }
1933
1934 /*
1935  * Make sure any pending truncation is applied before doing anything
1936  * that may depend on it.
1937  */
1938 void __ceph_do_pending_vmtruncate(struct inode *inode)
1939 {
1940         struct ceph_inode_info *ci = ceph_inode(inode);
1941         u64 to;
1942         int wrbuffer_refs, finish = 0;
1943
1944         mutex_lock(&ci->i_truncate_mutex);
1945 retry:
1946         spin_lock(&ci->i_ceph_lock);
1947         if (ci->i_truncate_pending == 0) {
1948                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1949                 spin_unlock(&ci->i_ceph_lock);
1950                 mutex_unlock(&ci->i_truncate_mutex);
1951                 return;
1952         }
1953
1954         /*
1955          * make sure any dirty snapped pages are flushed before we
1956          * possibly truncate them.. so write AND block!
1957          */
1958         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1959                 spin_unlock(&ci->i_ceph_lock);
1960                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1961                      inode);
1962                 filemap_write_and_wait_range(&inode->i_data, 0,
1963                                              inode->i_sb->s_maxbytes);
1964                 goto retry;
1965         }
1966
1967         /* there should be no reader or writer */
1968         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1969
1970         to = ci->i_truncate_size;
1971         wrbuffer_refs = ci->i_wrbuffer_ref;
1972         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1973              ci->i_truncate_pending, to);
1974         spin_unlock(&ci->i_ceph_lock);
1975
1976         truncate_pagecache(inode, to);
1977
1978         spin_lock(&ci->i_ceph_lock);
1979         if (to == ci->i_truncate_size) {
1980                 ci->i_truncate_pending = 0;
1981                 finish = 1;
1982         }
1983         spin_unlock(&ci->i_ceph_lock);
1984         if (!finish)
1985                 goto retry;
1986
1987         mutex_unlock(&ci->i_truncate_mutex);
1988
1989         if (wrbuffer_refs == 0)
1990                 ceph_check_caps(ci, 0, NULL);
1991
1992         wake_up_all(&ci->i_cap_wq);
1993 }
1994
1995 static void ceph_inode_work(struct work_struct *work)
1996 {
1997         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1998                                                  i_work);
1999         struct inode *inode = &ci->vfs_inode;
2000
2001         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2002                 dout("writeback %p\n", inode);
2003                 filemap_fdatawrite(&inode->i_data);
2004         }
2005         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2006                 ceph_do_invalidate_pages(inode);
2007
2008         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2009                 __ceph_do_pending_vmtruncate(inode);
2010
2011         iput(inode);
2012 }
2013
2014 /*
2015  * symlinks
2016  */
2017 static const struct inode_operations ceph_symlink_iops = {
2018         .get_link = simple_get_link,
2019         .setattr = ceph_setattr,
2020         .getattr = ceph_getattr,
2021         .listxattr = ceph_listxattr,
2022 };
2023
2024 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2025 {
2026         struct ceph_inode_info *ci = ceph_inode(inode);
2027         unsigned int ia_valid = attr->ia_valid;
2028         struct ceph_mds_request *req;
2029         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2030         struct ceph_cap_flush *prealloc_cf;
2031         int issued;
2032         int release = 0, dirtied = 0;
2033         int mask = 0;
2034         int err = 0;
2035         int inode_dirty_flags = 0;
2036         bool lock_snap_rwsem = false;
2037
2038         prealloc_cf = ceph_alloc_cap_flush();
2039         if (!prealloc_cf)
2040                 return -ENOMEM;
2041
2042         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2043                                        USE_AUTH_MDS);
2044         if (IS_ERR(req)) {
2045                 ceph_free_cap_flush(prealloc_cf);
2046                 return PTR_ERR(req);
2047         }
2048
2049         spin_lock(&ci->i_ceph_lock);
2050         issued = __ceph_caps_issued(ci, NULL);
2051
2052         if (!ci->i_head_snapc &&
2053             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2054                 lock_snap_rwsem = true;
2055                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2056                         spin_unlock(&ci->i_ceph_lock);
2057                         down_read(&mdsc->snap_rwsem);
2058                         spin_lock(&ci->i_ceph_lock);
2059                         issued = __ceph_caps_issued(ci, NULL);
2060                 }
2061         }
2062
2063         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2064
2065         if (ia_valid & ATTR_UID) {
2066                 dout("setattr %p uid %d -> %d\n", inode,
2067                      from_kuid(&init_user_ns, inode->i_uid),
2068                      from_kuid(&init_user_ns, attr->ia_uid));
2069                 if (issued & CEPH_CAP_AUTH_EXCL) {
2070                         inode->i_uid = attr->ia_uid;
2071                         dirtied |= CEPH_CAP_AUTH_EXCL;
2072                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2073                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2074                         req->r_args.setattr.uid = cpu_to_le32(
2075                                 from_kuid(&init_user_ns, attr->ia_uid));
2076                         mask |= CEPH_SETATTR_UID;
2077                         release |= CEPH_CAP_AUTH_SHARED;
2078                 }
2079         }
2080         if (ia_valid & ATTR_GID) {
2081                 dout("setattr %p gid %d -> %d\n", inode,
2082                      from_kgid(&init_user_ns, inode->i_gid),
2083                      from_kgid(&init_user_ns, attr->ia_gid));
2084                 if (issued & CEPH_CAP_AUTH_EXCL) {
2085                         inode->i_gid = attr->ia_gid;
2086                         dirtied |= CEPH_CAP_AUTH_EXCL;
2087                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2088                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2089                         req->r_args.setattr.gid = cpu_to_le32(
2090                                 from_kgid(&init_user_ns, attr->ia_gid));
2091                         mask |= CEPH_SETATTR_GID;
2092                         release |= CEPH_CAP_AUTH_SHARED;
2093                 }
2094         }
2095         if (ia_valid & ATTR_MODE) {
2096                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2097                      attr->ia_mode);
2098                 if (issued & CEPH_CAP_AUTH_EXCL) {
2099                         inode->i_mode = attr->ia_mode;
2100                         dirtied |= CEPH_CAP_AUTH_EXCL;
2101                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2102                            attr->ia_mode != inode->i_mode) {
2103                         inode->i_mode = attr->ia_mode;
2104                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2105                         mask |= CEPH_SETATTR_MODE;
2106                         release |= CEPH_CAP_AUTH_SHARED;
2107                 }
2108         }
2109
2110         if (ia_valid & ATTR_ATIME) {
2111                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2112                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2113                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2114                 if (issued & CEPH_CAP_FILE_EXCL) {
2115                         ci->i_time_warp_seq++;
2116                         inode->i_atime = attr->ia_atime;
2117                         dirtied |= CEPH_CAP_FILE_EXCL;
2118                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2119                            timespec64_compare(&inode->i_atime,
2120                                             &attr->ia_atime) < 0) {
2121                         inode->i_atime = attr->ia_atime;
2122                         dirtied |= CEPH_CAP_FILE_WR;
2123                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2124                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2125                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2126                                                &attr->ia_atime);
2127                         mask |= CEPH_SETATTR_ATIME;
2128                         release |= CEPH_CAP_FILE_SHARED |
2129                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2130                 }
2131         }
2132         if (ia_valid & ATTR_SIZE) {
2133                 dout("setattr %p size %lld -> %lld\n", inode,
2134                      inode->i_size, attr->ia_size);
2135                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2136                     attr->ia_size > inode->i_size) {
2137                         i_size_write(inode, attr->ia_size);
2138                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2139                         ci->i_reported_size = attr->ia_size;
2140                         dirtied |= CEPH_CAP_FILE_EXCL;
2141                         ia_valid |= ATTR_MTIME;
2142                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2143                            attr->ia_size != inode->i_size) {
2144                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2145                         req->r_args.setattr.old_size =
2146                                 cpu_to_le64(inode->i_size);
2147                         mask |= CEPH_SETATTR_SIZE;
2148                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2149                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2150                 }
2151         }
2152         if (ia_valid & ATTR_MTIME) {
2153                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2154                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2155                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2156                 if (issued & CEPH_CAP_FILE_EXCL) {
2157                         ci->i_time_warp_seq++;
2158                         inode->i_mtime = attr->ia_mtime;
2159                         dirtied |= CEPH_CAP_FILE_EXCL;
2160                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2161                            timespec64_compare(&inode->i_mtime,
2162                                             &attr->ia_mtime) < 0) {
2163                         inode->i_mtime = attr->ia_mtime;
2164                         dirtied |= CEPH_CAP_FILE_WR;
2165                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2166                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2167                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2168                                                &attr->ia_mtime);
2169                         mask |= CEPH_SETATTR_MTIME;
2170                         release |= CEPH_CAP_FILE_SHARED |
2171                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2172                 }
2173         }
2174
2175         /* these do nothing */
2176         if (ia_valid & ATTR_CTIME) {
2177                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2178                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2179                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2180                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2181                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2182                      only ? "ctime only" : "ignored");
2183                 if (only) {
2184                         /*
2185                          * if kernel wants to dirty ctime but nothing else,
2186                          * we need to choose a cap to dirty under, or do
2187                          * a almost-no-op setattr
2188                          */
2189                         if (issued & CEPH_CAP_AUTH_EXCL)
2190                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2191                         else if (issued & CEPH_CAP_FILE_EXCL)
2192                                 dirtied |= CEPH_CAP_FILE_EXCL;
2193                         else if (issued & CEPH_CAP_XATTR_EXCL)
2194                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2195                         else
2196                                 mask |= CEPH_SETATTR_CTIME;
2197                 }
2198         }
2199         if (ia_valid & ATTR_FILE)
2200                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2201
2202         if (dirtied) {
2203                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2204                                                            &prealloc_cf);
2205                 inode->i_ctime = attr->ia_ctime;
2206         }
2207
2208         release &= issued;
2209         spin_unlock(&ci->i_ceph_lock);
2210         if (lock_snap_rwsem)
2211                 up_read(&mdsc->snap_rwsem);
2212
2213         if (inode_dirty_flags)
2214                 __mark_inode_dirty(inode, inode_dirty_flags);
2215
2216
2217         if (mask) {
2218                 req->r_inode = inode;
2219                 ihold(inode);
2220                 req->r_inode_drop = release;
2221                 req->r_args.setattr.mask = cpu_to_le32(mask);
2222                 req->r_num_caps = 1;
2223                 req->r_stamp = attr->ia_ctime;
2224                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2225         }
2226         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2227              ceph_cap_string(dirtied), mask);
2228
2229         ceph_mdsc_put_request(req);
2230         ceph_free_cap_flush(prealloc_cf);
2231
2232         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2233                 __ceph_do_pending_vmtruncate(inode);
2234
2235         return err;
2236 }
2237
2238 /*
2239  * setattr
2240  */
2241 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2242 {
2243         struct inode *inode = d_inode(dentry);
2244         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2245         int err;
2246
2247         if (ceph_snap(inode) != CEPH_NOSNAP)
2248                 return -EROFS;
2249
2250         err = setattr_prepare(dentry, attr);
2251         if (err != 0)
2252                 return err;
2253
2254         if ((attr->ia_valid & ATTR_SIZE) &&
2255             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2256                 return -EFBIG;
2257
2258         if ((attr->ia_valid & ATTR_SIZE) &&
2259             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2260                 return -EDQUOT;
2261
2262         err = __ceph_setattr(inode, attr);
2263
2264         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2265                 err = posix_acl_chmod(inode, attr->ia_mode);
2266
2267         return err;
2268 }
2269
2270 /*
2271  * Verify that we have a lease on the given mask.  If not,
2272  * do a getattr against an mds.
2273  */
2274 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2275                       int mask, bool force)
2276 {
2277         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2278         struct ceph_mds_client *mdsc = fsc->mdsc;
2279         struct ceph_mds_request *req;
2280         int mode;
2281         int err;
2282
2283         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2284                 dout("do_getattr inode %p SNAPDIR\n", inode);
2285                 return 0;
2286         }
2287
2288         dout("do_getattr inode %p mask %s mode 0%o\n",
2289              inode, ceph_cap_string(mask), inode->i_mode);
2290         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2291                         return 0;
2292
2293         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2294         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2295         if (IS_ERR(req))
2296                 return PTR_ERR(req);
2297         req->r_inode = inode;
2298         ihold(inode);
2299         req->r_num_caps = 1;
2300         req->r_args.getattr.mask = cpu_to_le32(mask);
2301         req->r_locked_page = locked_page;
2302         err = ceph_mdsc_do_request(mdsc, NULL, req);
2303         if (locked_page && err == 0) {
2304                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2305                 if (inline_version == 0) {
2306                         /* the reply is supposed to contain inline data */
2307                         err = -EINVAL;
2308                 } else if (inline_version == CEPH_INLINE_NONE) {
2309                         err = -ENODATA;
2310                 } else {
2311                         err = req->r_reply_info.targeti.inline_len;
2312                 }
2313         }
2314         ceph_mdsc_put_request(req);
2315         dout("do_getattr result=%d\n", err);
2316         return err;
2317 }
2318
2319
2320 /*
2321  * Check inode permissions.  We verify we have a valid value for
2322  * the AUTH cap, then call the generic handler.
2323  */
2324 int ceph_permission(struct inode *inode, int mask)
2325 {
2326         int err;
2327
2328         if (mask & MAY_NOT_BLOCK)
2329                 return -ECHILD;
2330
2331         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2332
2333         if (!err)
2334                 err = generic_permission(inode, mask);
2335         return err;
2336 }
2337
2338 /* Craft a mask of needed caps given a set of requested statx attrs. */
2339 static int statx_to_caps(u32 want)
2340 {
2341         int mask = 0;
2342
2343         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2344                 mask |= CEPH_CAP_AUTH_SHARED;
2345
2346         if (want & (STATX_NLINK|STATX_CTIME))
2347                 mask |= CEPH_CAP_LINK_SHARED;
2348
2349         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2350                     STATX_BLOCKS))
2351                 mask |= CEPH_CAP_FILE_SHARED;
2352
2353         if (want & (STATX_CTIME))
2354                 mask |= CEPH_CAP_XATTR_SHARED;
2355
2356         return mask;
2357 }
2358
2359 /*
2360  * Get all the attributes. If we have sufficient caps for the requested attrs,
2361  * then we can avoid talking to the MDS at all.
2362  */
2363 int ceph_getattr(const struct path *path, struct kstat *stat,
2364                  u32 request_mask, unsigned int flags)
2365 {
2366         struct inode *inode = d_inode(path->dentry);
2367         struct ceph_inode_info *ci = ceph_inode(inode);
2368         u32 valid_mask = STATX_BASIC_STATS;
2369         int err = 0;
2370
2371         /* Skip the getattr altogether if we're asked not to sync */
2372         if (!(flags & AT_STATX_DONT_SYNC)) {
2373                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2374                                       flags & AT_STATX_FORCE_SYNC);
2375                 if (err)
2376                         return err;
2377         }
2378
2379         generic_fillattr(inode, stat);
2380         stat->ino = ceph_present_inode(inode);
2381
2382         /*
2383          * btime on newly-allocated inodes is 0, so if this is still set to
2384          * that, then assume that it's not valid.
2385          */
2386         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2387                 stat->btime = ci->i_btime;
2388                 valid_mask |= STATX_BTIME;
2389         }
2390
2391         if (ceph_snap(inode) == CEPH_NOSNAP)
2392                 stat->dev = inode->i_sb->s_dev;
2393         else
2394                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2395
2396         if (S_ISDIR(inode->i_mode)) {
2397                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2398                                         RBYTES))
2399                         stat->size = ci->i_rbytes;
2400                 else
2401                         stat->size = ci->i_files + ci->i_subdirs;
2402                 stat->blocks = 0;
2403                 stat->blksize = 65536;
2404                 /*
2405                  * Some applications rely on the number of st_nlink
2406                  * value on directories to be either 0 (if unlinked)
2407                  * or 2 + number of subdirectories.
2408                  */
2409                 if (stat->nlink == 1)
2410                         /* '.' + '..' + subdirs */
2411                         stat->nlink = 1 + 1 + ci->i_subdirs;
2412         }
2413
2414         stat->result_mask = request_mask & valid_mask;
2415         return err;
2416 }