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         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
  46
  47         ci->i_vino = *(struct ceph_vino *)data;
  48         inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
  49         inode_set_iversion_raw(inode, 0);
  50         percpu_counter_inc(&mdsc->metric.total_inodes);
  51
  52         return 0;
  53 }
  54
  55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  56 {
  57         struct inode *inode;
  58
  59         if (ceph_vino_is_reserved(vino))
  60                 return ERR_PTR(-EREMOTEIO);
  61
  62         inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
  63                              ceph_set_ino_cb, &vino);
  64         if (!inode)
  65                 return ERR_PTR(-ENOMEM);
  66
  67         dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
  68              ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
  69         return inode;
  70 }
  71
  72 /*
  73  * get/constuct snapdir inode for a given directory
  74  */
  75 struct inode *ceph_get_snapdir(struct inode *parent)
  76 {
  77         struct ceph_vino vino = {
  78                 .ino = ceph_ino(parent),
  79                 .snap = CEPH_SNAPDIR,
  80         };
  81         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  82         struct ceph_inode_info *ci = ceph_inode(inode);
  83
  84         BUG_ON(!S_ISDIR(parent->i_mode));
  85         if (IS_ERR(inode))
  86                 return inode;
  87         inode->i_mode = parent->i_mode;
  88         inode->i_uid = parent->i_uid;
  89         inode->i_gid = parent->i_gid;
  90         inode->i_mtime = parent->i_mtime;
  91         inode->i_ctime = parent->i_ctime;
  92         inode->i_atime = parent->i_atime;
  93         ci->i_rbytes = 0;
  94         ci->i_btime = ceph_inode(parent)->i_btime;
  95
  96         if (inode->i_state & I_NEW) {
  97                 inode->i_op = &ceph_snapdir_iops;
  98                 inode->i_fop = &ceph_snapdir_fops;
  99                 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
 100                 unlock_new_inode(inode);
 101         }
 102
 103         return inode;
 104 }
 105
 106 const struct inode_operations ceph_file_iops = {
 107         .permission = ceph_permission,
 108         .setattr = ceph_setattr,
 109         .getattr = ceph_getattr,
 110         .listxattr = ceph_listxattr,
 111         .get_acl = ceph_get_acl,
 112         .set_acl = ceph_set_acl,
 113 };
 114
 115
 116 /*
 117  * We use a 'frag tree' to keep track of the MDS's directory fragments
 118  * for a given inode (usually there is just a single fragment).  We
 119  * need to know when a child frag is delegated to a new MDS, or when
 120  * it is flagged as replicated, so we can direct our requests
 121  * accordingly.
 122  */
 123
 124 /*
 125  * find/create a frag in the tree
 126  */
 127 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 128                                                     u32 f)
 129 {
 130         struct rb_node **p;
 131         struct rb_node *parent = NULL;
 132         struct ceph_inode_frag *frag;
 133         int c;
 134
 135         p = &ci->i_fragtree.rb_node;
 136         while (*p) {
 137                 parent = *p;
 138                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 139                 c = ceph_frag_compare(f, frag->frag);
 140                 if (c < 0)
 141                         p = &(*p)->rb_left;
 142                 else if (c > 0)
 143                         p = &(*p)->rb_right;
 144                 else
 145                         return frag;
 146         }
 147
 148         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 149         if (!frag)
 150                 return ERR_PTR(-ENOMEM);
 151
 152         frag->frag = f;
 153         frag->split_by = 0;
 154         frag->mds = -1;
 155         frag->ndist = 0;
 156
 157         rb_link_node(&frag->node, parent, p);
 158         rb_insert_color(&frag->node, &ci->i_fragtree);
 159
 160         dout("get_or_create_frag added %llx.%llx frag %x\n",
 161              ceph_vinop(&ci->vfs_inode), f);
 162         return frag;
 163 }
 164
 165 /*
 166  * find a specific frag @f
 167  */
 168 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 169 {
 170         struct rb_node *n = ci->i_fragtree.rb_node;
 171
 172         while (n) {
 173                 struct ceph_inode_frag *frag =
 174                         rb_entry(n, struct ceph_inode_frag, node);
 175                 int c = ceph_frag_compare(f, frag->frag);
 176                 if (c < 0)
 177                         n = n->rb_left;
 178                 else if (c > 0)
 179                         n = n->rb_right;
 180                 else
 181                         return frag;
 182         }
 183         return NULL;
 184 }
 185
 186 /*
 187  * Choose frag containing the given value @v.  If @pfrag is
 188  * specified, copy the frag delegation info to the caller if
 189  * it is present.
 190  */
 191 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 192                               struct ceph_inode_frag *pfrag, int *found)
 193 {
 194         u32 t = ceph_frag_make(0, 0);
 195         struct ceph_inode_frag *frag;
 196         unsigned nway, i;
 197         u32 n;
 198
 199         if (found)
 200                 *found = 0;
 201
 202         while (1) {
 203                 WARN_ON(!ceph_frag_contains_value(t, v));
 204                 frag = __ceph_find_frag(ci, t);
 205                 if (!frag)
 206                         break; /* t is a leaf */
 207                 if (frag->split_by == 0) {
 208                         if (pfrag)
 209                                 memcpy(pfrag, frag, sizeof(*pfrag));
 210                         if (found)
 211                                 *found = 1;
 212                         break;
 213                 }
 214
 215                 /* choose child */
 216                 nway = 1 << frag->split_by;
 217                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 218                      frag->split_by, nway);
 219                 for (i = 0; i < nway; i++) {
 220                         n = ceph_frag_make_child(t, frag->split_by, i);
 221                         if (ceph_frag_contains_value(n, v)) {
 222                                 t = n;
 223                                 break;
 224                         }
 225                 }
 226                 BUG_ON(i == nway);
 227         }
 228         dout("choose_frag(%x) = %x\n", v, t);
 229
 230         return t;
 231 }
 232
 233 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 234                      struct ceph_inode_frag *pfrag, int *found)
 235 {
 236         u32 ret;
 237         mutex_lock(&ci->i_fragtree_mutex);
 238         ret = __ceph_choose_frag(ci, v, pfrag, found);
 239         mutex_unlock(&ci->i_fragtree_mutex);
 240         return ret;
 241 }
 242
 243 /*
 244  * Process dirfrag (delegation) info from the mds.  Include leaf
 245  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 246  * branches/splits are included in i_fragtree)
 247  */
 248 static int ceph_fill_dirfrag(struct inode *inode,
 249                              struct ceph_mds_reply_dirfrag *dirinfo)
 250 {
 251         struct ceph_inode_info *ci = ceph_inode(inode);
 252         struct ceph_inode_frag *frag;
 253         u32 id = le32_to_cpu(dirinfo->frag);
 254         int mds = le32_to_cpu(dirinfo->auth);
 255         int ndist = le32_to_cpu(dirinfo->ndist);
 256         int diri_auth = -1;
 257         int i;
 258         int err = 0;
 259
 260         spin_lock(&ci->i_ceph_lock);
 261         if (ci->i_auth_cap)
 262                 diri_auth = ci->i_auth_cap->mds;
 263         spin_unlock(&ci->i_ceph_lock);
 264
 265         if (mds == -1) /* CDIR_AUTH_PARENT */
 266                 mds = diri_auth;
 267
 268         mutex_lock(&ci->i_fragtree_mutex);
 269         if (ndist == 0 && mds == diri_auth) {
 270                 /* no delegation info needed. */
 271                 frag = __ceph_find_frag(ci, id);
 272                 if (!frag)
 273                         goto out;
 274                 if (frag->split_by == 0) {
 275                         /* tree leaf, remove */
 276                         dout("fill_dirfrag removed %llx.%llx frag %x"
 277                              " (no ref)\n", ceph_vinop(inode), id);
 278                         rb_erase(&frag->node, &ci->i_fragtree);
 279                         kfree(frag);
 280                 } else {
 281                         /* tree branch, keep and clear */
 282                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 283                              " referral\n", ceph_vinop(inode), id);
 284                         frag->mds = -1;
 285                         frag->ndist = 0;
 286                 }
 287                 goto out;
 288         }
 289
 290
 291         /* find/add this frag to store mds delegation info */
 292         frag = __get_or_create_frag(ci, id);
 293         if (IS_ERR(frag)) {
 294                 /* this is not the end of the world; we can continue
 295                    with bad/inaccurate delegation info */
 296                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 297                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 298                 err = -ENOMEM;
 299                 goto out;
 300         }
 301
 302         frag->mds = mds;
 303         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 304         for (i = 0; i < frag->ndist; i++)
 305                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 306         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 307              ceph_vinop(inode), frag->frag, frag->ndist);
 308
 309 out:
 310         mutex_unlock(&ci->i_fragtree_mutex);
 311         return err;
 312 }
 313
 314 static int frag_tree_split_cmp(const void *l, const void *r)
 315 {
 316         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 317         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 318         return ceph_frag_compare(le32_to_cpu(ls->frag),
 319                                  le32_to_cpu(rs->frag));
 320 }
 321
 322 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 323 {
 324         if (!frag)
 325                 return f == ceph_frag_make(0, 0);
 326         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 327                 return false;
 328         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 329 }
 330
 331 static int ceph_fill_fragtree(struct inode *inode,
 332                               struct ceph_frag_tree_head *fragtree,
 333                               struct ceph_mds_reply_dirfrag *dirinfo)
 334 {
 335         struct ceph_inode_info *ci = ceph_inode(inode);
 336         struct ceph_inode_frag *frag, *prev_frag = NULL;
 337         struct rb_node *rb_node;
 338         unsigned i, split_by, nsplits;
 339         u32 id;
 340         bool update = false;
 341
 342         mutex_lock(&ci->i_fragtree_mutex);
 343         nsplits = le32_to_cpu(fragtree->nsplits);
 344         if (nsplits != ci->i_fragtree_nsplits) {
 345                 update = true;
 346         } else if (nsplits) {
 347                 i = prandom_u32() % nsplits;
 348                 id = le32_to_cpu(fragtree->splits[i].frag);
 349                 if (!__ceph_find_frag(ci, id))
 350                         update = true;
 351         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 352                 rb_node = rb_first(&ci->i_fragtree);
 353                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 354                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 355                         update = true;
 356         }
 357         if (!update && dirinfo) {
 358                 id = le32_to_cpu(dirinfo->frag);
 359                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 360                         update = true;
 361         }
 362         if (!update)
 363                 goto out_unlock;
 364
 365         if (nsplits > 1) {
 366                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 367                      frag_tree_split_cmp, NULL);
 368         }
 369
 370         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 371         rb_node = rb_first(&ci->i_fragtree);
 372         for (i = 0; i < nsplits; i++) {
 373                 id = le32_to_cpu(fragtree->splits[i].frag);
 374                 split_by = le32_to_cpu(fragtree->splits[i].by);
 375                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 376                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 377                                "frag %x split by %d\n", ceph_vinop(inode),
 378                                i, nsplits, id, split_by);
 379                         continue;
 380                 }
 381                 frag = NULL;
 382                 while (rb_node) {
 383                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 384                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 385                                 if (frag->frag != id)
 386                                         frag = NULL;
 387                                 else
 388                                         rb_node = rb_next(rb_node);
 389                                 break;
 390                         }
 391                         rb_node = rb_next(rb_node);
 392                         /* delete stale split/leaf node */
 393                         if (frag->split_by > 0 ||
 394                             !is_frag_child(frag->frag, prev_frag)) {
 395                                 rb_erase(&frag->node, &ci->i_fragtree);
 396                                 if (frag->split_by > 0)
 397                                         ci->i_fragtree_nsplits--;
 398                                 kfree(frag);
 399                         }
 400                         frag = NULL;
 401                 }
 402                 if (!frag) {
 403                         frag = __get_or_create_frag(ci, id);
 404                         if (IS_ERR(frag))
 405                                 continue;
 406                 }
 407                 if (frag->split_by == 0)
 408                         ci->i_fragtree_nsplits++;
 409                 frag->split_by = split_by;
 410                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 411                 prev_frag = frag;
 412         }
 413         while (rb_node) {
 414                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 415                 rb_node = rb_next(rb_node);
 416                 /* delete stale split/leaf node */
 417                 if (frag->split_by > 0 ||
 418                     !is_frag_child(frag->frag, prev_frag)) {
 419                         rb_erase(&frag->node, &ci->i_fragtree);
 420                         if (frag->split_by > 0)
 421                                 ci->i_fragtree_nsplits--;
 422                         kfree(frag);
 423                 }
 424         }
 425 out_unlock:
 426         mutex_unlock(&ci->i_fragtree_mutex);
 427         return 0;
 428 }
 429
 430 /*
 431  * initialize a newly allocated inode.
 432  */
 433 struct inode *ceph_alloc_inode(struct super_block *sb)
 434 {
 435         struct ceph_inode_info *ci;
 436         int i;
 437
 438         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 439         if (!ci)
 440                 return NULL;
 441
 442         dout("alloc_inode %p\n", &ci->vfs_inode);
 443
 444         spin_lock_init(&ci->i_ceph_lock);
 445
 446         ci->i_version = 0;
 447         ci->i_inline_version = 0;
 448         ci->i_time_warp_seq = 0;
 449         ci->i_ceph_flags = 0;
 450         atomic64_set(&ci->i_ordered_count, 1);
 451         atomic64_set(&ci->i_release_count, 1);
 452         atomic64_set(&ci->i_complete_seq[0], 0);
 453         atomic64_set(&ci->i_complete_seq[1], 0);
 454         ci->i_symlink = NULL;
 455
 456         ci->i_max_bytes = 0;
 457         ci->i_max_files = 0;
 458
 459         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 460         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 461         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 462
 463         ci->i_fragtree = RB_ROOT;
 464         mutex_init(&ci->i_fragtree_mutex);
 465
 466         ci->i_xattrs.blob = NULL;
 467         ci->i_xattrs.prealloc_blob = NULL;
 468         ci->i_xattrs.dirty = false;
 469         ci->i_xattrs.index = RB_ROOT;
 470         ci->i_xattrs.count = 0;
 471         ci->i_xattrs.names_size = 0;
 472         ci->i_xattrs.vals_size = 0;
 473         ci->i_xattrs.version = 0;
 474         ci->i_xattrs.index_version = 0;
 475
 476         ci->i_caps = RB_ROOT;
 477         ci->i_auth_cap = NULL;
 478         ci->i_dirty_caps = 0;
 479         ci->i_flushing_caps = 0;
 480         INIT_LIST_HEAD(&ci->i_dirty_item);
 481         INIT_LIST_HEAD(&ci->i_flushing_item);
 482         ci->i_prealloc_cap_flush = NULL;
 483         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 484         init_waitqueue_head(&ci->i_cap_wq);
 485         ci->i_hold_caps_max = 0;
 486         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 487         INIT_LIST_HEAD(&ci->i_cap_snaps);
 488         ci->i_head_snapc = NULL;
 489         ci->i_snap_caps = 0;
 490
 491         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 492         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 493                 ci->i_nr_by_mode[i] = 0;
 494
 495         mutex_init(&ci->i_truncate_mutex);
 496         ci->i_truncate_seq = 0;
 497         ci->i_truncate_size = 0;
 498         ci->i_truncate_pending = 0;
 499
 500         ci->i_max_size = 0;
 501         ci->i_reported_size = 0;
 502         ci->i_wanted_max_size = 0;
 503         ci->i_requested_max_size = 0;
 504
 505         ci->i_pin_ref = 0;
 506         ci->i_rd_ref = 0;
 507         ci->i_rdcache_ref = 0;
 508         ci->i_wr_ref = 0;
 509         ci->i_wb_ref = 0;
 510         ci->i_fx_ref = 0;
 511         ci->i_wrbuffer_ref = 0;
 512         ci->i_wrbuffer_ref_head = 0;
 513         atomic_set(&ci->i_filelock_ref, 0);
 514         atomic_set(&ci->i_shared_gen, 1);
 515         ci->i_rdcache_gen = 0;
 516         ci->i_rdcache_revoking = 0;
 517
 518         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 519         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 520         spin_lock_init(&ci->i_unsafe_lock);
 521
 522         ci->i_snap_realm = NULL;
 523         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 524         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 525
 526         INIT_WORK(&ci->i_work, ceph_inode_work);
 527         ci->i_work_mask = 0;
 528         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 529
 530         ceph_fscache_inode_init(ci);
 531
 532         return &ci->vfs_inode;
 533 }
 534
 535 void ceph_free_inode(struct inode *inode)
 536 {
 537         struct ceph_inode_info *ci = ceph_inode(inode);
 538
 539         kfree(ci->i_symlink);
 540         kmem_cache_free(ceph_inode_cachep, ci);
 541 }
 542
 543 void ceph_evict_inode(struct inode *inode)
 544 {
 545         struct ceph_inode_info *ci = ceph_inode(inode);
 546         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 547         struct ceph_inode_frag *frag;
 548         struct rb_node *n;
 549
 550         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 551
 552         percpu_counter_dec(&mdsc->metric.total_inodes);
 553
 554         truncate_inode_pages_final(&inode->i_data);
 555         clear_inode(inode);
 556
 557         ceph_fscache_unregister_inode_cookie(ci);
 558
 559         __ceph_remove_caps(ci);
 560
 561         if (__ceph_has_any_quota(ci))
 562                 ceph_adjust_quota_realms_count(inode, false);
 563
 564         /*
 565          * we may still have a snap_realm reference if there are stray
 566          * caps in i_snap_caps.
 567          */
 568         if (ci->i_snap_realm) {
 569                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 570                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 571                         dout(" dropping residual ref to snap realm %p\n",
 572                              realm);
 573                         spin_lock(&realm->inodes_with_caps_lock);
 574                         list_del_init(&ci->i_snap_realm_item);
 575                         ci->i_snap_realm = NULL;
 576                         if (realm->ino == ci->i_vino.ino)
 577                                 realm->inode = NULL;
 578                         spin_unlock(&realm->inodes_with_caps_lock);
 579                         ceph_put_snap_realm(mdsc, realm);
 580                 } else {
 581                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 582                         ci->i_snap_realm = NULL;
 583                 }
 584         }
 585
 586         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 587                 frag = rb_entry(n, struct ceph_inode_frag, node);
 588                 rb_erase(n, &ci->i_fragtree);
 589                 kfree(frag);
 590         }
 591         ci->i_fragtree_nsplits = 0;
 592
 593         __ceph_destroy_xattrs(ci);
 594         if (ci->i_xattrs.blob)
 595                 ceph_buffer_put(ci->i_xattrs.blob);
 596         if (ci->i_xattrs.prealloc_blob)
 597                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 598
 599         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 600         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 601 }
 602
 603 static inline blkcnt_t calc_inode_blocks(u64 size)
 604 {
 605         return (size + (1<<9) - 1) >> 9;
 606 }
 607
 608 /*
 609  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 610  * careful because either the client or MDS may have more up to date
 611  * info, depending on which capabilities are held, and whether
 612  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 613  * and size are monotonically increasing, except when utimes() or
 614  * truncate() increments the corresponding _seq values.)
 615  */
 616 int ceph_fill_file_size(struct inode *inode, int issued,
 617                         u32 truncate_seq, u64 truncate_size, u64 size)
 618 {
 619         struct ceph_inode_info *ci = ceph_inode(inode);
 620         int queue_trunc = 0;
 621
 622         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 623             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 624                 dout("size %lld -> %llu\n", inode->i_size, size);
 625                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 626                         pr_err("fill_file_size non-zero size for directory\n");
 627                         size = 0;
 628                 }
 629                 i_size_write(inode, size);
 630                 inode->i_blocks = calc_inode_blocks(size);
 631                 ci->i_reported_size = size;
 632                 if (truncate_seq != ci->i_truncate_seq) {
 633                         dout("truncate_seq %u -> %u\n",
 634                              ci->i_truncate_seq, truncate_seq);
 635                         ci->i_truncate_seq = truncate_seq;
 636
 637                         /* the MDS should have revoked these caps */
 638                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_RD |
 639                                                CEPH_CAP_FILE_LAZYIO));
 640                         /*
 641                          * If we hold relevant caps, or in the case where we're
 642                          * not the only client referencing this file and we
 643                          * don't hold those caps, then we need to check whether
 644                          * the file is either opened or mmaped
 645                          */
 646                         if ((issued & (CEPH_CAP_FILE_CACHE|
 647                                        CEPH_CAP_FILE_BUFFER)) ||
 648                             mapping_mapped(inode->i_mapping) ||
 649                             __ceph_is_file_opened(ci)) {
 650                                 ci->i_truncate_pending++;
 651                                 queue_trunc = 1;
 652                         }
 653                 }
 654         }
 655         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 656             ci->i_truncate_size != truncate_size) {
 657                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 658                      truncate_size);
 659                 ci->i_truncate_size = truncate_size;
 660         }
 661
 662         if (queue_trunc)
 663                 ceph_fscache_invalidate(inode);
 664
 665         return queue_trunc;
 666 }
 667
 668 void ceph_fill_file_time(struct inode *inode, int issued,
 669                          u64 time_warp_seq, struct timespec64 *ctime,
 670                          struct timespec64 *mtime, struct timespec64 *atime)
 671 {
 672         struct ceph_inode_info *ci = ceph_inode(inode);
 673         int warn = 0;
 674
 675         if (issued & (CEPH_CAP_FILE_EXCL|
 676                       CEPH_CAP_FILE_WR|
 677                       CEPH_CAP_FILE_BUFFER|
 678                       CEPH_CAP_AUTH_EXCL|
 679                       CEPH_CAP_XATTR_EXCL)) {
 680                 if (ci->i_version == 0 ||
 681                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 682                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 683                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 684                              ctime->tv_sec, ctime->tv_nsec);
 685                         inode->i_ctime = *ctime;
 686                 }
 687                 if (ci->i_version == 0 ||
 688                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 689                         /* the MDS did a utimes() */
 690                         dout("mtime %lld.%09ld -> %lld.%09ld "
 691                              "tw %d -> %d\n",
 692                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 693                              mtime->tv_sec, mtime->tv_nsec,
 694                              ci->i_time_warp_seq, (int)time_warp_seq);
 695
 696                         inode->i_mtime = *mtime;
 697                         inode->i_atime = *atime;
 698                         ci->i_time_warp_seq = time_warp_seq;
 699                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 700                         /* nobody did utimes(); take the max */
 701                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 702                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 703                                      inode->i_mtime.tv_sec,
 704                                      inode->i_mtime.tv_nsec,
 705                                      mtime->tv_sec, mtime->tv_nsec);
 706                                 inode->i_mtime = *mtime;
 707                         }
 708                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 709                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 710                                      inode->i_atime.tv_sec,
 711                                      inode->i_atime.tv_nsec,
 712                                      atime->tv_sec, atime->tv_nsec);
 713                                 inode->i_atime = *atime;
 714                         }
 715                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 716                         /* we did a utimes(); ignore mds values */
 717                 } else {
 718                         warn = 1;
 719                 }
 720         } else {
 721                 /* we have no write|excl caps; whatever the MDS says is true */
 722                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 723                         inode->i_ctime = *ctime;
 724                         inode->i_mtime = *mtime;
 725                         inode->i_atime = *atime;
 726                         ci->i_time_warp_seq = time_warp_seq;
 727                 } else {
 728                         warn = 1;
 729                 }
 730         }
 731         if (warn) /* time_warp_seq shouldn't go backwards */
 732                 dout("%p mds time_warp_seq %llu < %u\n",
 733                      inode, time_warp_seq, ci->i_time_warp_seq);
 734 }
 735
 736 /*
 737  * Populate an inode based on info from mds.  May be called on new or
 738  * existing inodes.
 739  */
 740 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 741                     struct ceph_mds_reply_info_in *iinfo,
 742                     struct ceph_mds_reply_dirfrag *dirinfo,
 743                     struct ceph_mds_session *session, int cap_fmode,
 744                     struct ceph_cap_reservation *caps_reservation)
 745 {
 746         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 747         struct ceph_mds_reply_inode *info = iinfo->in;
 748         struct ceph_inode_info *ci = ceph_inode(inode);
 749         int issued, new_issued, info_caps;
 750         struct timespec64 mtime, atime, ctime;
 751         struct ceph_buffer *xattr_blob = NULL;
 752         struct ceph_buffer *old_blob = NULL;
 753         struct ceph_string *pool_ns = NULL;
 754         struct ceph_cap *new_cap = NULL;
 755         int err = 0;
 756         bool wake = false;
 757         bool queue_trunc = false;
 758         bool new_version = false;
 759         bool fill_inline = false;
 760
 761         lockdep_assert_held(&mdsc->snap_rwsem);
 762
 763         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 764              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 765              ci->i_version);
 766
 767         info_caps = le32_to_cpu(info->cap.caps);
 768
 769         /* prealloc new cap struct */
 770         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 771                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 772                 if (!new_cap)
 773                         return -ENOMEM;
 774         }
 775
 776         /*
 777          * prealloc xattr data, if it looks like we'll need it.  only
 778          * if len > 4 (meaning there are actually xattrs; the first 4
 779          * bytes are the xattr count).
 780          */
 781         if (iinfo->xattr_len > 4) {
 782                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 783                 if (!xattr_blob)
 784                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 785                                iinfo->xattr_len);
 786         }
 787
 788         if (iinfo->pool_ns_len > 0)
 789                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 790                                                      iinfo->pool_ns_len);
 791
 792         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 793                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 794
 795         spin_lock(&ci->i_ceph_lock);
 796
 797         /*
 798          * provided version will be odd if inode value is projected,
 799          * even if stable.  skip the update if we have newer stable
 800          * info (ours>=theirs, e.g. due to racing mds replies), unless
 801          * we are getting projected (unstable) info (in which case the
 802          * version is odd, and we want ours>theirs).
 803          *   us   them
 804          *   2    2     skip
 805          *   3    2     skip
 806          *   3    3     update
 807          */
 808         if (ci->i_version == 0 ||
 809             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 810              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 811                 new_version = true;
 812
 813         /* Update change_attribute */
 814         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 815
 816         __ceph_caps_issued(ci, &issued);
 817         issued |= __ceph_caps_dirty(ci);
 818         new_issued = ~issued & info_caps;
 819
 820         /* update inode */
 821         inode->i_rdev = le32_to_cpu(info->rdev);
 822         /* directories have fl_stripe_unit set to zero */
 823         if (le32_to_cpu(info->layout.fl_stripe_unit))
 824                 inode->i_blkbits =
 825                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 826         else
 827                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 828
 829         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 830
 831         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 832             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 833                 inode->i_mode = le32_to_cpu(info->mode);
 834                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 835                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 836                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 837                      from_kuid(&init_user_ns, inode->i_uid),
 838                      from_kgid(&init_user_ns, inode->i_gid));
 839                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 840                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 841         }
 842
 843         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 844             (issued & CEPH_CAP_LINK_EXCL) == 0)
 845                 set_nlink(inode, le32_to_cpu(info->nlink));
 846
 847         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 848                 /* be careful with mtime, atime, size */
 849                 ceph_decode_timespec64(&atime, &info->atime);
 850                 ceph_decode_timespec64(&mtime, &info->mtime);
 851                 ceph_decode_timespec64(&ctime, &info->ctime);
 852                 ceph_fill_file_time(inode, issued,
 853                                 le32_to_cpu(info->time_warp_seq),
 854                                 &ctime, &mtime, &atime);
 855         }
 856
 857         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 858                 ci->i_files = le64_to_cpu(info->files);
 859                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 860         }
 861
 862         if (new_version ||
 863             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 864                 s64 old_pool = ci->i_layout.pool_id;
 865                 struct ceph_string *old_ns;
 866
 867                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 868                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 869                                         lockdep_is_held(&ci->i_ceph_lock));
 870                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 871
 872                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 873                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 874
 875                 pool_ns = old_ns;
 876
 877                 queue_trunc = ceph_fill_file_size(inode, issued,
 878                                         le32_to_cpu(info->truncate_seq),
 879                                         le64_to_cpu(info->truncate_size),
 880                                         le64_to_cpu(info->size));
 881                 /* only update max_size on auth cap */
 882                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 883                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 884                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 885                                         le64_to_cpu(info->max_size));
 886                         ci->i_max_size = le64_to_cpu(info->max_size);
 887                 }
 888         }
 889
 890         /* layout and rstat are not tracked by capability, update them if
 891          * the inode info is from auth mds */
 892         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 893                 if (S_ISDIR(inode->i_mode)) {
 894                         ci->i_dir_layout = iinfo->dir_layout;
 895                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 896                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 897                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 898                         ci->i_dir_pin = iinfo->dir_pin;
 899                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 900                 }
 901         }
 902
 903         /* xattrs */
 904         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 905         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 906             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 907                 if (ci->i_xattrs.blob)
 908                         old_blob = ci->i_xattrs.blob;
 909                 ci->i_xattrs.blob = xattr_blob;
 910                 if (xattr_blob)
 911                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 912                                iinfo->xattr_data, iinfo->xattr_len);
 913                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 914                 ceph_forget_all_cached_acls(inode);
 915                 ceph_security_invalidate_secctx(inode);
 916                 xattr_blob = NULL;
 917         }
 918
 919         /* finally update i_version */
 920         if (le64_to_cpu(info->version) > ci->i_version)
 921                 ci->i_version = le64_to_cpu(info->version);
 922
 923         inode->i_mapping->a_ops = &ceph_aops;
 924
 925         switch (inode->i_mode & S_IFMT) {
 926         case S_IFIFO:
 927         case S_IFBLK:
 928         case S_IFCHR:
 929         case S_IFSOCK:
 930                 inode->i_blkbits = PAGE_SHIFT;
 931                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 932                 inode->i_op = &ceph_file_iops;
 933                 break;
 934         case S_IFREG:
 935                 inode->i_op = &ceph_file_iops;
 936                 inode->i_fop = &ceph_file_fops;
 937                 break;
 938         case S_IFLNK:
 939                 inode->i_op = &ceph_symlink_iops;
 940                 if (!ci->i_symlink) {
 941                         u32 symlen = iinfo->symlink_len;
 942                         char *sym;
 943
 944                         spin_unlock(&ci->i_ceph_lock);
 945
 946                         if (symlen != i_size_read(inode)) {
 947                                 pr_err("%s %llx.%llx BAD symlink "
 948                                         "size %lld\n", __func__,
 949                                         ceph_vinop(inode),
 950                                         i_size_read(inode));
 951                                 i_size_write(inode, symlen);
 952                                 inode->i_blocks = calc_inode_blocks(symlen);
 953                         }
 954
 955                         err = -ENOMEM;
 956                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 957                         if (!sym)
 958                                 goto out;
 959
 960                         spin_lock(&ci->i_ceph_lock);
 961                         if (!ci->i_symlink)
 962                                 ci->i_symlink = sym;
 963                         else
 964                                 kfree(sym); /* lost a race */
 965                 }
 966                 inode->i_link = ci->i_symlink;
 967                 break;
 968         case S_IFDIR:
 969                 inode->i_op = &ceph_dir_iops;
 970                 inode->i_fop = &ceph_dir_fops;
 971                 break;
 972         default:
 973                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
 974                        ceph_vinop(inode), inode->i_mode);
 975         }
 976
 977         /* were we issued a capability? */
 978         if (info_caps) {
 979                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 980                         ceph_add_cap(inode, session,
 981                                      le64_to_cpu(info->cap.cap_id),
 982                                      info_caps,
 983                                      le32_to_cpu(info->cap.wanted),
 984                                      le32_to_cpu(info->cap.seq),
 985                                      le32_to_cpu(info->cap.mseq),
 986                                      le64_to_cpu(info->cap.realm),
 987                                      info->cap.flags, &new_cap);
 988
 989                         /* set dir completion flag? */
 990                         if (S_ISDIR(inode->i_mode) &&
 991                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 992                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 993                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 994                             !__ceph_dir_is_complete(ci)) {
 995                                 dout(" marking %p complete (empty)\n", inode);
 996                                 i_size_write(inode, 0);
 997                                 __ceph_dir_set_complete(ci,
 998                                         atomic64_read(&ci->i_release_count),
 999                                         atomic64_read(&ci->i_ordered_count));
1000                         }
1001
1002                         wake = true;
1003                 } else {
1004                         dout(" %p got snap_caps %s\n", inode,
1005                              ceph_cap_string(info_caps));
1006                         ci->i_snap_caps |= info_caps;
1007                 }
1008         }
1009
1010         if (iinfo->inline_version > 0 &&
1011             iinfo->inline_version >= ci->i_inline_version) {
1012                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1013                 ci->i_inline_version = iinfo->inline_version;
1014                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1015                     (locked_page || (info_caps & cache_caps)))
1016                         fill_inline = true;
1017         }
1018
1019         if (cap_fmode >= 0) {
1020                 if (!info_caps)
1021                         pr_warn("mds issued no caps on %llx.%llx\n",
1022                                 ceph_vinop(inode));
1023                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1024         }
1025
1026         spin_unlock(&ci->i_ceph_lock);
1027
1028         if (fill_inline)
1029                 ceph_fill_inline_data(inode, locked_page,
1030                                       iinfo->inline_data, iinfo->inline_len);
1031
1032         if (wake)
1033                 wake_up_all(&ci->i_cap_wq);
1034
1035         /* queue truncate if we saw i_size decrease */
1036         if (queue_trunc)
1037                 ceph_queue_vmtruncate(inode);
1038
1039         /* populate frag tree */
1040         if (S_ISDIR(inode->i_mode))
1041                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1042
1043         /* update delegation info? */
1044         if (dirinfo)
1045                 ceph_fill_dirfrag(inode, dirinfo);
1046
1047         err = 0;
1048 out:
1049         if (new_cap)
1050                 ceph_put_cap(mdsc, new_cap);
1051         ceph_buffer_put(old_blob);
1052         ceph_buffer_put(xattr_blob);
1053         ceph_put_string(pool_ns);
1054         return err;
1055 }
1056
1057 /*
1058  * caller should hold session s_mutex and dentry->d_lock.
1059  */
1060 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1061                                   struct ceph_mds_reply_lease *lease,
1062                                   struct ceph_mds_session *session,
1063                                   unsigned long from_time,
1064                                   struct ceph_mds_session **old_lease_session)
1065 {
1066         struct ceph_dentry_info *di = ceph_dentry(dentry);
1067         unsigned mask = le16_to_cpu(lease->mask);
1068         long unsigned duration = le32_to_cpu(lease->duration_ms);
1069         long unsigned ttl = from_time + (duration * HZ) / 1000;
1070         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1071
1072         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1073              dentry, duration, ttl);
1074
1075         /* only track leases on regular dentries */
1076         if (ceph_snap(dir) != CEPH_NOSNAP)
1077                 return;
1078
1079         if (mask & CEPH_LEASE_PRIMARY_LINK)
1080                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1081         else
1082                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1083
1084         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1085         if (!(mask & CEPH_LEASE_VALID)) {
1086                 __ceph_dentry_dir_lease_touch(di);
1087                 return;
1088         }
1089
1090         if (di->lease_gen == session->s_cap_gen &&
1091             time_before(ttl, di->time))
1092                 return;  /* we already have a newer lease. */
1093
1094         if (di->lease_session && di->lease_session != session) {
1095                 *old_lease_session = di->lease_session;
1096                 di->lease_session = NULL;
1097         }
1098
1099         if (!di->lease_session)
1100                 di->lease_session = ceph_get_mds_session(session);
1101         di->lease_gen = session->s_cap_gen;
1102         di->lease_seq = le32_to_cpu(lease->seq);
1103         di->lease_renew_after = half_ttl;
1104         di->lease_renew_from = 0;
1105         di->time = ttl;
1106
1107         __ceph_dentry_lease_touch(di);
1108 }
1109
1110 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1111                                         struct ceph_mds_reply_lease *lease,
1112                                         struct ceph_mds_session *session,
1113                                         unsigned long from_time)
1114 {
1115         struct ceph_mds_session *old_lease_session = NULL;
1116         spin_lock(&dentry->d_lock);
1117         __update_dentry_lease(dir, dentry, lease, session, from_time,
1118                               &old_lease_session);
1119         spin_unlock(&dentry->d_lock);
1120         ceph_put_mds_session(old_lease_session);
1121 }
1122
1123 /*
1124  * update dentry lease without having parent inode locked
1125  */
1126 static void update_dentry_lease_careful(struct dentry *dentry,
1127                                         struct ceph_mds_reply_lease *lease,
1128                                         struct ceph_mds_session *session,
1129                                         unsigned long from_time,
1130                                         char *dname, u32 dname_len,
1131                                         struct ceph_vino *pdvino,
1132                                         struct ceph_vino *ptvino)
1133
1134 {
1135         struct inode *dir;
1136         struct ceph_mds_session *old_lease_session = NULL;
1137
1138         spin_lock(&dentry->d_lock);
1139         /* make sure dentry's name matches target */
1140         if (dentry->d_name.len != dname_len ||
1141             memcmp(dentry->d_name.name, dname, dname_len))
1142                 goto out_unlock;
1143
1144         dir = d_inode(dentry->d_parent);
1145         /* make sure parent matches dvino */
1146         if (!ceph_ino_compare(dir, pdvino))
1147                 goto out_unlock;
1148
1149         /* make sure dentry's inode matches target. NULL ptvino means that
1150          * we expect a negative dentry */
1151         if (ptvino) {
1152                 if (d_really_is_negative(dentry))
1153                         goto out_unlock;
1154                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1155                         goto out_unlock;
1156         } else {
1157                 if (d_really_is_positive(dentry))
1158                         goto out_unlock;
1159         }
1160
1161         __update_dentry_lease(dir, dentry, lease, session,
1162                               from_time, &old_lease_session);
1163 out_unlock:
1164         spin_unlock(&dentry->d_lock);
1165         ceph_put_mds_session(old_lease_session);
1166 }
1167
1168 /*
1169  * splice a dentry to an inode.
1170  * caller must hold directory i_mutex for this to be safe.
1171  */
1172 static int splice_dentry(struct dentry **pdn, struct inode *in)
1173 {
1174         struct dentry *dn = *pdn;
1175         struct dentry *realdn;
1176
1177         BUG_ON(d_inode(dn));
1178
1179         if (S_ISDIR(in->i_mode)) {
1180                 /* If inode is directory, d_splice_alias() below will remove
1181                  * 'realdn' from its origin parent. We need to ensure that
1182                  * origin parent's readdir cache will not reference 'realdn'
1183                  */
1184                 realdn = d_find_any_alias(in);
1185                 if (realdn) {
1186                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1187                         spin_lock(&realdn->d_lock);
1188
1189                         realdn->d_op->d_prune(realdn);
1190
1191                         di->time = jiffies;
1192                         di->lease_shared_gen = 0;
1193                         di->offset = 0;
1194
1195                         spin_unlock(&realdn->d_lock);
1196                         dput(realdn);
1197                 }
1198         }
1199
1200         /* dn must be unhashed */
1201         if (!d_unhashed(dn))
1202                 d_drop(dn);
1203         realdn = d_splice_alias(in, dn);
1204         if (IS_ERR(realdn)) {
1205                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1206                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1207                 return PTR_ERR(realdn);
1208         }
1209
1210         if (realdn) {
1211                 dout("dn %p (%d) spliced with %p (%d) "
1212                      "inode %p ino %llx.%llx\n",
1213                      dn, d_count(dn),
1214                      realdn, d_count(realdn),
1215                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1216                 dput(dn);
1217                 *pdn = realdn;
1218         } else {
1219                 BUG_ON(!ceph_dentry(dn));
1220                 dout("dn %p attached to %p ino %llx.%llx\n",
1221                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1222         }
1223         return 0;
1224 }
1225
1226 /*
1227  * Incorporate results into the local cache.  This is either just
1228  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1229  * after a lookup).
1230  *
1231  * A reply may contain
1232  *         a directory inode along with a dentry.
1233  *  and/or a target inode
1234  *
1235  * Called with snap_rwsem (read).
1236  */
1237 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1238 {
1239         struct ceph_mds_session *session = req->r_session;
1240         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1241         struct inode *in = NULL;
1242         struct ceph_vino tvino, dvino;
1243         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1244         int err = 0;
1245
1246         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1247              rinfo->head->is_dentry, rinfo->head->is_target);
1248
1249         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1250                 dout("fill_trace reply is empty!\n");
1251                 if (rinfo->head->result == 0 && req->r_parent)
1252                         ceph_invalidate_dir_request(req);
1253                 return 0;
1254         }
1255
1256         if (rinfo->head->is_dentry) {
1257                 struct inode *dir = req->r_parent;
1258
1259                 if (dir) {
1260                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1261                                               rinfo->dirfrag, session, -1,
1262                                               &req->r_caps_reservation);
1263                         if (err < 0)
1264                                 goto done;
1265                 } else {
1266                         WARN_ON_ONCE(1);
1267                 }
1268
1269                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1270                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1271                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1272                         struct qstr dname;
1273                         struct dentry *dn, *parent;
1274
1275                         BUG_ON(!rinfo->head->is_target);
1276                         BUG_ON(req->r_dentry);
1277
1278                         parent = d_find_any_alias(dir);
1279                         BUG_ON(!parent);
1280
1281                         dname.name = rinfo->dname;
1282                         dname.len = rinfo->dname_len;
1283                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1284                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1285                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1286 retry_lookup:
1287                         dn = d_lookup(parent, &dname);
1288                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1289                              parent, dname.len, dname.name, dn);
1290
1291                         if (!dn) {
1292                                 dn = d_alloc(parent, &dname);
1293                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1294                                      dname.len, dname.name, dn);
1295                                 if (!dn) {
1296                                         dput(parent);
1297                                         err = -ENOMEM;
1298                                         goto done;
1299                                 }
1300                                 err = 0;
1301                         } else if (d_really_is_positive(dn) &&
1302                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1303                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1304                                 dout(" dn %p points to wrong inode %p\n",
1305                                      dn, d_inode(dn));
1306                                 ceph_dir_clear_ordered(dir);
1307                                 d_delete(dn);
1308                                 dput(dn);
1309                                 goto retry_lookup;
1310                         }
1311
1312                         req->r_dentry = dn;
1313                         dput(parent);
1314                 }
1315         }
1316
1317         if (rinfo->head->is_target) {
1318                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1319                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1320
1321                 in = ceph_get_inode(sb, tvino);
1322                 if (IS_ERR(in)) {
1323                         err = PTR_ERR(in);
1324                         goto done;
1325                 }
1326
1327                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1328                                 NULL, session,
1329                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1330                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1331                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1332                                 &req->r_caps_reservation);
1333                 if (err < 0) {
1334                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1335                                 in, ceph_vinop(in));
1336                         if (in->i_state & I_NEW)
1337                                 discard_new_inode(in);
1338                         else
1339                                 iput(in);
1340                         goto done;
1341                 }
1342                 req->r_target_inode = in;
1343                 if (in->i_state & I_NEW)
1344                         unlock_new_inode(in);
1345         }
1346
1347         /*
1348          * ignore null lease/binding on snapdir ENOENT, or else we
1349          * will have trouble splicing in the virtual snapdir later
1350          */
1351         if (rinfo->head->is_dentry &&
1352             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1353             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1354             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1355                                                fsc->mount_options->snapdir_name,
1356                                                req->r_dentry->d_name.len))) {
1357                 /*
1358                  * lookup link rename   : null -> possibly existing inode
1359                  * mknod symlink mkdir  : null -> new inode
1360                  * unlink               : linked -> null
1361                  */
1362                 struct inode *dir = req->r_parent;
1363                 struct dentry *dn = req->r_dentry;
1364                 bool have_dir_cap, have_lease;
1365
1366                 BUG_ON(!dn);
1367                 BUG_ON(!dir);
1368                 BUG_ON(d_inode(dn->d_parent) != dir);
1369
1370                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1371                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1372
1373                 BUG_ON(ceph_ino(dir) != dvino.ino);
1374                 BUG_ON(ceph_snap(dir) != dvino.snap);
1375
1376                 /* do we have a lease on the whole dir? */
1377                 have_dir_cap =
1378                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1379                          CEPH_CAP_FILE_SHARED);
1380
1381                 /* do we have a dn lease? */
1382                 have_lease = have_dir_cap ||
1383                         le32_to_cpu(rinfo->dlease->duration_ms);
1384                 if (!have_lease)
1385                         dout("fill_trace  no dentry lease or dir cap\n");
1386
1387                 /* rename? */
1388                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1389                         struct inode *olddir = req->r_old_dentry_dir;
1390                         BUG_ON(!olddir);
1391
1392                         dout(" src %p '%pd' dst %p '%pd'\n",
1393                              req->r_old_dentry,
1394                              req->r_old_dentry,
1395                              dn, dn);
1396                         dout("fill_trace doing d_move %p -> %p\n",
1397                              req->r_old_dentry, dn);
1398
1399                         /* d_move screws up sibling dentries' offsets */
1400                         ceph_dir_clear_ordered(dir);
1401                         ceph_dir_clear_ordered(olddir);
1402
1403                         d_move(req->r_old_dentry, dn);
1404                         dout(" src %p '%pd' dst %p '%pd'\n",
1405                              req->r_old_dentry,
1406                              req->r_old_dentry,
1407                              dn, dn);
1408
1409                         /* ensure target dentry is invalidated, despite
1410                            rehashing bug in vfs_rename_dir */
1411                         ceph_invalidate_dentry_lease(dn);
1412
1413                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1414                              ceph_dentry(req->r_old_dentry)->offset);
1415
1416                         /* swap r_dentry and r_old_dentry in case that
1417                          * splice_dentry() gets called later. This is safe
1418                          * because no other place will use them */
1419                         req->r_dentry = req->r_old_dentry;
1420                         req->r_old_dentry = dn;
1421                         dn = req->r_dentry;
1422                 }
1423
1424                 /* null dentry? */
1425                 if (!rinfo->head->is_target) {
1426                         dout("fill_trace null dentry\n");
1427                         if (d_really_is_positive(dn)) {
1428                                 dout("d_delete %p\n", dn);
1429                                 ceph_dir_clear_ordered(dir);
1430                                 d_delete(dn);
1431                         } else if (have_lease) {
1432                                 if (d_unhashed(dn))
1433                                         d_add(dn, NULL);
1434                                 update_dentry_lease(dir, dn,
1435                                                     rinfo->dlease, session,
1436                                                     req->r_request_started);
1437                         }
1438                         goto done;
1439                 }
1440
1441                 /* attach proper inode */
1442                 if (d_really_is_negative(dn)) {
1443                         ceph_dir_clear_ordered(dir);
1444                         ihold(in);
1445                         err = splice_dentry(&req->r_dentry, in);
1446                         if (err < 0)
1447                                 goto done;
1448                         dn = req->r_dentry;  /* may have spliced */
1449                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1450                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1451                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1452                              ceph_vinop(in));
1453                         d_invalidate(dn);
1454                         have_lease = false;
1455                 }
1456
1457                 if (have_lease) {
1458                         update_dentry_lease(dir, dn,
1459                                             rinfo->dlease, session,
1460                                             req->r_request_started);
1461                 }
1462                 dout(" final dn %p\n", dn);
1463         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1464                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1465                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1466                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1467                 struct inode *dir = req->r_parent;
1468
1469                 /* fill out a snapdir LOOKUPSNAP dentry */
1470                 BUG_ON(!dir);
1471                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1472                 BUG_ON(!req->r_dentry);
1473                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1474                 ceph_dir_clear_ordered(dir);
1475                 ihold(in);
1476                 err = splice_dentry(&req->r_dentry, in);
1477                 if (err < 0)
1478                         goto done;
1479         } else if (rinfo->head->is_dentry && req->r_dentry) {
1480                 /* parent inode is not locked, be carefull */
1481                 struct ceph_vino *ptvino = NULL;
1482                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1483                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1484                 if (rinfo->head->is_target) {
1485                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1486                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1487                         ptvino = &tvino;
1488                 }
1489                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1490                                             session, req->r_request_started,
1491                                             rinfo->dname, rinfo->dname_len,
1492                                             &dvino, ptvino);
1493         }
1494 done:
1495         dout("fill_trace done err=%d\n", err);
1496         return err;
1497 }
1498
1499 /*
1500  * Prepopulate our cache with readdir results, leases, etc.
1501  */
1502 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1503                                            struct ceph_mds_session *session)
1504 {
1505         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1506         int i, err = 0;
1507
1508         for (i = 0; i < rinfo->dir_nr; i++) {
1509                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1510                 struct ceph_vino vino;
1511                 struct inode *in;
1512                 int rc;
1513
1514                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1515                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1516
1517                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1518                 if (IS_ERR(in)) {
1519                         err = PTR_ERR(in);
1520                         dout("new_inode badness got %d\n", err);
1521                         continue;
1522                 }
1523                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1524                                      -1, &req->r_caps_reservation);
1525                 if (rc < 0) {
1526                         pr_err("ceph_fill_inode badness on %p got %d\n",
1527                                in, rc);
1528                         err = rc;
1529                         if (in->i_state & I_NEW) {
1530                                 ihold(in);
1531                                 discard_new_inode(in);
1532                         }
1533                 } else if (in->i_state & I_NEW) {
1534                         unlock_new_inode(in);
1535                 }
1536
1537                 /* avoid calling iput_final() in mds dispatch threads */
1538                 ceph_async_iput(in);
1539         }
1540
1541         return err;
1542 }
1543
1544 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1545 {
1546         if (ctl->page) {
1547                 kunmap(ctl->page);
1548                 put_page(ctl->page);
1549                 ctl->page = NULL;
1550         }
1551 }
1552
1553 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1554                               struct ceph_readdir_cache_control *ctl,
1555                               struct ceph_mds_request *req)
1556 {
1557         struct ceph_inode_info *ci = ceph_inode(dir);
1558         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1559         unsigned idx = ctl->index % nsize;
1560         pgoff_t pgoff = ctl->index / nsize;
1561
1562         if (!ctl->page || pgoff != page_index(ctl->page)) {
1563                 ceph_readdir_cache_release(ctl);
1564                 if (idx == 0)
1565                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1566                 else
1567                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1568                 if (!ctl->page) {
1569                         ctl->index = -1;
1570                         return idx == 0 ? -ENOMEM : 0;
1571                 }
1572                 /* reading/filling the cache are serialized by
1573                  * i_mutex, no need to use page lock */
1574                 unlock_page(ctl->page);
1575                 ctl->dentries = kmap(ctl->page);
1576                 if (idx == 0)
1577                         memset(ctl->dentries, 0, PAGE_SIZE);
1578         }
1579
1580         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1581             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1582                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1583                 ctl->dentries[idx] = dn;
1584                 ctl->index++;
1585         } else {
1586                 dout("disable readdir cache\n");
1587                 ctl->index = -1;
1588         }
1589         return 0;
1590 }
1591
1592 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1593                              struct ceph_mds_session *session)
1594 {
1595         struct dentry *parent = req->r_dentry;
1596         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1597         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1598         struct qstr dname;
1599         struct dentry *dn;
1600         struct inode *in;
1601         int err = 0, skipped = 0, ret, i;
1602         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1603         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1604         u32 last_hash = 0;
1605         u32 fpos_offset;
1606         struct ceph_readdir_cache_control cache_ctl = {};
1607
1608         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1609                 return readdir_prepopulate_inodes_only(req, session);
1610
1611         if (rinfo->hash_order) {
1612                 if (req->r_path2) {
1613                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1614                                                   req->r_path2,
1615                                                   strlen(req->r_path2));
1616                         last_hash = ceph_frag_value(last_hash);
1617                 } else if (rinfo->offset_hash) {
1618                         /* mds understands offset_hash */
1619                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1620                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1621                 }
1622         }
1623
1624         if (rinfo->dir_dir &&
1625             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1626                 dout("readdir_prepopulate got new frag %x -> %x\n",
1627                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1628                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1629                 if (!rinfo->hash_order)
1630                         req->r_readdir_offset = 2;
1631         }
1632
1633         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1634                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1635                      rinfo->dir_nr, parent);
1636         } else {
1637                 dout("readdir_prepopulate %d items under dn %p\n",
1638                      rinfo->dir_nr, parent);
1639                 if (rinfo->dir_dir)
1640                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1641
1642                 if (ceph_frag_is_leftmost(frag) &&
1643                     req->r_readdir_offset == 2 &&
1644                     !(rinfo->hash_order && last_hash)) {
1645                         /* note dir version at start of readdir so we can
1646                          * tell if any dentries get dropped */
1647                         req->r_dir_release_cnt =
1648                                 atomic64_read(&ci->i_release_count);
1649                         req->r_dir_ordered_cnt =
1650                                 atomic64_read(&ci->i_ordered_count);
1651                         req->r_readdir_cache_idx = 0;
1652                 }
1653         }
1654
1655         cache_ctl.index = req->r_readdir_cache_idx;
1656         fpos_offset = req->r_readdir_offset;
1657
1658         /* FIXME: release caps/leases if error occurs */
1659         for (i = 0; i < rinfo->dir_nr; i++) {
1660                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1661                 struct ceph_vino tvino;
1662
1663                 dname.name = rde->name;
1664                 dname.len = rde->name_len;
1665                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1666
1667                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1668                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1669
1670                 if (rinfo->hash_order) {
1671                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1672                                                  rde->name, rde->name_len);
1673                         hash = ceph_frag_value(hash);
1674                         if (hash != last_hash)
1675                                 fpos_offset = 2;
1676                         last_hash = hash;
1677                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1678                 } else {
1679                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1680                 }
1681
1682 retry_lookup:
1683                 dn = d_lookup(parent, &dname);
1684                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1685                      parent, dname.len, dname.name, dn);
1686
1687                 if (!dn) {
1688                         dn = d_alloc(parent, &dname);
1689                         dout("d_alloc %p '%.*s' = %p\n", parent,
1690                              dname.len, dname.name, dn);
1691                         if (!dn) {
1692                                 dout("d_alloc badness\n");
1693                                 err = -ENOMEM;
1694                                 goto out;
1695                         }
1696                 } else if (d_really_is_positive(dn) &&
1697                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1698                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1699                         struct ceph_dentry_info *di = ceph_dentry(dn);
1700                         dout(" dn %p points to wrong inode %p\n",
1701                              dn, d_inode(dn));
1702
1703                         spin_lock(&dn->d_lock);
1704                         if (di->offset > 0 &&
1705                             di->lease_shared_gen ==
1706                             atomic_read(&ci->i_shared_gen)) {
1707                                 __ceph_dir_clear_ordered(ci);
1708                                 di->offset = 0;
1709                         }
1710                         spin_unlock(&dn->d_lock);
1711
1712                         d_delete(dn);
1713                         dput(dn);
1714                         goto retry_lookup;
1715                 }
1716
1717                 /* inode */
1718                 if (d_really_is_positive(dn)) {
1719                         in = d_inode(dn);
1720                 } else {
1721                         in = ceph_get_inode(parent->d_sb, tvino);
1722                         if (IS_ERR(in)) {
1723                                 dout("new_inode badness\n");
1724                                 d_drop(dn);
1725                                 dput(dn);
1726                                 err = PTR_ERR(in);
1727                                 goto out;
1728                         }
1729                 }
1730
1731                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1732                                       -1, &req->r_caps_reservation);
1733                 if (ret < 0) {
1734                         pr_err("ceph_fill_inode badness on %p\n", in);
1735                         if (d_really_is_negative(dn)) {
1736                                 /* avoid calling iput_final() in mds
1737                                  * dispatch threads */
1738                                 if (in->i_state & I_NEW) {
1739                                         ihold(in);
1740                                         discard_new_inode(in);
1741                                 }
1742                                 ceph_async_iput(in);
1743                         }
1744                         d_drop(dn);
1745                         err = ret;
1746                         goto next_item;
1747                 }
1748                 if (in->i_state & I_NEW)
1749                         unlock_new_inode(in);
1750
1751                 if (d_really_is_negative(dn)) {
1752                         if (ceph_security_xattr_deadlock(in)) {
1753                                 dout(" skip splicing dn %p to inode %p"
1754                                      " (security xattr deadlock)\n", dn, in);
1755                                 ceph_async_iput(in);
1756                                 skipped++;
1757                                 goto next_item;
1758                         }
1759
1760                         err = splice_dentry(&dn, in);
1761                         if (err < 0)
1762                                 goto next_item;
1763                 }
1764
1765                 ceph_dentry(dn)->offset = rde->offset;
1766
1767                 update_dentry_lease(d_inode(parent), dn,
1768                                     rde->lease, req->r_session,
1769                                     req->r_request_started);
1770
1771                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1772                         ret = fill_readdir_cache(d_inode(parent), dn,
1773                                                  &cache_ctl, req);
1774                         if (ret < 0)
1775                                 err = ret;
1776                 }
1777 next_item:
1778                 dput(dn);
1779         }
1780 out:
1781         if (err == 0 && skipped == 0) {
1782                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1783                 req->r_readdir_cache_idx = cache_ctl.index;
1784         }
1785         ceph_readdir_cache_release(&cache_ctl);
1786         dout("readdir_prepopulate done\n");
1787         return err;
1788 }
1789
1790 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1791 {
1792         struct ceph_inode_info *ci = ceph_inode(inode);
1793         bool ret;
1794
1795         spin_lock(&ci->i_ceph_lock);
1796         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1797         i_size_write(inode, size);
1798         inode->i_blocks = calc_inode_blocks(size);
1799
1800         ret = __ceph_should_report_size(ci);
1801
1802         spin_unlock(&ci->i_ceph_lock);
1803         return ret;
1804 }
1805
1806 /*
1807  * Put reference to inode, but avoid calling iput_final() in current thread.
1808  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1809  * some contexts.
1810  */
1811 void ceph_async_iput(struct inode *inode)
1812 {
1813         if (!inode)
1814                 return;
1815         for (;;) {
1816                 if (atomic_add_unless(&inode->i_count, -1, 1))
1817                         break;
1818                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1819                                &ceph_inode(inode)->i_work))
1820                         break;
1821                 /* queue work failed, i_count must be at least 2 */
1822         }
1823 }
1824
1825 /*
1826  * Write back inode data in a worker thread.  (This can't be done
1827  * in the message handler context.)
1828  */
1829 void ceph_queue_writeback(struct inode *inode)
1830 {
1831         struct ceph_inode_info *ci = ceph_inode(inode);
1832         set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1833
1834         ihold(inode);
1835         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1836                        &ci->i_work)) {
1837                 dout("ceph_queue_writeback %p\n", inode);
1838         } else {
1839                 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1840                      inode, ci->i_work_mask);
1841                 iput(inode);
1842         }
1843 }
1844
1845 /*
1846  * queue an async invalidation
1847  */
1848 void ceph_queue_invalidate(struct inode *inode)
1849 {
1850         struct ceph_inode_info *ci = ceph_inode(inode);
1851         set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1852
1853         ihold(inode);
1854         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1855                        &ceph_inode(inode)->i_work)) {
1856                 dout("ceph_queue_invalidate %p\n", inode);
1857         } else {
1858                 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1859                      inode, ci->i_work_mask);
1860                 iput(inode);
1861         }
1862 }
1863
1864 /*
1865  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1866  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1867  */
1868 void ceph_queue_vmtruncate(struct inode *inode)
1869 {
1870         struct ceph_inode_info *ci = ceph_inode(inode);
1871         set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1872
1873         ihold(inode);
1874         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1875                        &ci->i_work)) {
1876                 dout("ceph_queue_vmtruncate %p\n", inode);
1877         } else {
1878                 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1879                      inode, ci->i_work_mask);
1880                 iput(inode);
1881         }
1882 }
1883
1884 static void ceph_do_invalidate_pages(struct inode *inode)
1885 {
1886         struct ceph_inode_info *ci = ceph_inode(inode);
1887         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1888         u32 orig_gen;
1889         int check = 0;
1890
1891         mutex_lock(&ci->i_truncate_mutex);
1892
1893         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1894                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1895                                     inode, ceph_ino(inode));
1896                 mapping_set_error(inode->i_mapping, -EIO);
1897                 truncate_pagecache(inode, 0);
1898                 mutex_unlock(&ci->i_truncate_mutex);
1899                 goto out;
1900         }
1901
1902         spin_lock(&ci->i_ceph_lock);
1903         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1904              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1905         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1906                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1907                         check = 1;
1908                 spin_unlock(&ci->i_ceph_lock);
1909                 mutex_unlock(&ci->i_truncate_mutex);
1910                 goto out;
1911         }
1912         orig_gen = ci->i_rdcache_gen;
1913         spin_unlock(&ci->i_ceph_lock);
1914
1915         ceph_fscache_invalidate(inode);
1916         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1917                 pr_err("invalidate_pages %p fails\n", inode);
1918         }
1919
1920         spin_lock(&ci->i_ceph_lock);
1921         if (orig_gen == ci->i_rdcache_gen &&
1922             orig_gen == ci->i_rdcache_revoking) {
1923                 dout("invalidate_pages %p gen %d successful\n", inode,
1924                      ci->i_rdcache_gen);
1925                 ci->i_rdcache_revoking--;
1926                 check = 1;
1927         } else {
1928                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1929                      inode, orig_gen, ci->i_rdcache_gen,
1930                      ci->i_rdcache_revoking);
1931                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1932                         check = 1;
1933         }
1934         spin_unlock(&ci->i_ceph_lock);
1935         mutex_unlock(&ci->i_truncate_mutex);
1936 out:
1937         if (check)
1938                 ceph_check_caps(ci, 0, NULL);
1939 }
1940
1941 /*
1942  * Make sure any pending truncation is applied before doing anything
1943  * that may depend on it.
1944  */
1945 void __ceph_do_pending_vmtruncate(struct inode *inode)
1946 {
1947         struct ceph_inode_info *ci = ceph_inode(inode);
1948         u64 to;
1949         int wrbuffer_refs, finish = 0;
1950
1951         mutex_lock(&ci->i_truncate_mutex);
1952 retry:
1953         spin_lock(&ci->i_ceph_lock);
1954         if (ci->i_truncate_pending == 0) {
1955                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1956                 spin_unlock(&ci->i_ceph_lock);
1957                 mutex_unlock(&ci->i_truncate_mutex);
1958                 return;
1959         }
1960
1961         /*
1962          * make sure any dirty snapped pages are flushed before we
1963          * possibly truncate them.. so write AND block!
1964          */
1965         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1966                 spin_unlock(&ci->i_ceph_lock);
1967                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1968                      inode);
1969                 filemap_write_and_wait_range(&inode->i_data, 0,
1970                                              inode->i_sb->s_maxbytes);
1971                 goto retry;
1972         }
1973
1974         /* there should be no reader or writer */
1975         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1976
1977         to = ci->i_truncate_size;
1978         wrbuffer_refs = ci->i_wrbuffer_ref;
1979         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1980              ci->i_truncate_pending, to);
1981         spin_unlock(&ci->i_ceph_lock);
1982
1983         truncate_pagecache(inode, to);
1984
1985         spin_lock(&ci->i_ceph_lock);
1986         if (to == ci->i_truncate_size) {
1987                 ci->i_truncate_pending = 0;
1988                 finish = 1;
1989         }
1990         spin_unlock(&ci->i_ceph_lock);
1991         if (!finish)
1992                 goto retry;
1993
1994         mutex_unlock(&ci->i_truncate_mutex);
1995
1996         if (wrbuffer_refs == 0)
1997                 ceph_check_caps(ci, 0, NULL);
1998
1999         wake_up_all(&ci->i_cap_wq);
2000 }
2001
2002 static void ceph_inode_work(struct work_struct *work)
2003 {
2004         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2005                                                  i_work);
2006         struct inode *inode = &ci->vfs_inode;
2007
2008         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2009                 dout("writeback %p\n", inode);
2010                 filemap_fdatawrite(&inode->i_data);
2011         }
2012         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2013                 ceph_do_invalidate_pages(inode);
2014
2015         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2016                 __ceph_do_pending_vmtruncate(inode);
2017
2018         iput(inode);
2019 }
2020
2021 /*
2022  * symlinks
2023  */
2024 static const struct inode_operations ceph_symlink_iops = {
2025         .get_link = simple_get_link,
2026         .setattr = ceph_setattr,
2027         .getattr = ceph_getattr,
2028         .listxattr = ceph_listxattr,
2029 };
2030
2031 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2032 {
2033         struct ceph_inode_info *ci = ceph_inode(inode);
2034         unsigned int ia_valid = attr->ia_valid;
2035         struct ceph_mds_request *req;
2036         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2037         struct ceph_cap_flush *prealloc_cf;
2038         int issued;
2039         int release = 0, dirtied = 0;
2040         int mask = 0;
2041         int err = 0;
2042         int inode_dirty_flags = 0;
2043         bool lock_snap_rwsem = false;
2044
2045         prealloc_cf = ceph_alloc_cap_flush();
2046         if (!prealloc_cf)
2047                 return -ENOMEM;
2048
2049         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2050                                        USE_AUTH_MDS);
2051         if (IS_ERR(req)) {
2052                 ceph_free_cap_flush(prealloc_cf);
2053                 return PTR_ERR(req);
2054         }
2055
2056         spin_lock(&ci->i_ceph_lock);
2057         issued = __ceph_caps_issued(ci, NULL);
2058
2059         if (!ci->i_head_snapc &&
2060             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2061                 lock_snap_rwsem = true;
2062                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2063                         spin_unlock(&ci->i_ceph_lock);
2064                         down_read(&mdsc->snap_rwsem);
2065                         spin_lock(&ci->i_ceph_lock);
2066                         issued = __ceph_caps_issued(ci, NULL);
2067                 }
2068         }
2069
2070         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2071
2072         if (ia_valid & ATTR_UID) {
2073                 dout("setattr %p uid %d -> %d\n", inode,
2074                      from_kuid(&init_user_ns, inode->i_uid),
2075                      from_kuid(&init_user_ns, attr->ia_uid));
2076                 if (issued & CEPH_CAP_AUTH_EXCL) {
2077                         inode->i_uid = attr->ia_uid;
2078                         dirtied |= CEPH_CAP_AUTH_EXCL;
2079                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2080                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2081                         req->r_args.setattr.uid = cpu_to_le32(
2082                                 from_kuid(&init_user_ns, attr->ia_uid));
2083                         mask |= CEPH_SETATTR_UID;
2084                         release |= CEPH_CAP_AUTH_SHARED;
2085                 }
2086         }
2087         if (ia_valid & ATTR_GID) {
2088                 dout("setattr %p gid %d -> %d\n", inode,
2089                      from_kgid(&init_user_ns, inode->i_gid),
2090                      from_kgid(&init_user_ns, attr->ia_gid));
2091                 if (issued & CEPH_CAP_AUTH_EXCL) {
2092                         inode->i_gid = attr->ia_gid;
2093                         dirtied |= CEPH_CAP_AUTH_EXCL;
2094                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2095                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2096                         req->r_args.setattr.gid = cpu_to_le32(
2097                                 from_kgid(&init_user_ns, attr->ia_gid));
2098                         mask |= CEPH_SETATTR_GID;
2099                         release |= CEPH_CAP_AUTH_SHARED;
2100                 }
2101         }
2102         if (ia_valid & ATTR_MODE) {
2103                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2104                      attr->ia_mode);
2105                 if (issued & CEPH_CAP_AUTH_EXCL) {
2106                         inode->i_mode = attr->ia_mode;
2107                         dirtied |= CEPH_CAP_AUTH_EXCL;
2108                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2109                            attr->ia_mode != inode->i_mode) {
2110                         inode->i_mode = attr->ia_mode;
2111                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2112                         mask |= CEPH_SETATTR_MODE;
2113                         release |= CEPH_CAP_AUTH_SHARED;
2114                 }
2115         }
2116
2117         if (ia_valid & ATTR_ATIME) {
2118                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2119                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2120                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2121                 if (issued & CEPH_CAP_FILE_EXCL) {
2122                         ci->i_time_warp_seq++;
2123                         inode->i_atime = attr->ia_atime;
2124                         dirtied |= CEPH_CAP_FILE_EXCL;
2125                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2126                            timespec64_compare(&inode->i_atime,
2127                                             &attr->ia_atime) < 0) {
2128                         inode->i_atime = attr->ia_atime;
2129                         dirtied |= CEPH_CAP_FILE_WR;
2130                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2131                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2132                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2133                                                &attr->ia_atime);
2134                         mask |= CEPH_SETATTR_ATIME;
2135                         release |= CEPH_CAP_FILE_SHARED |
2136                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2137                 }
2138         }
2139         if (ia_valid & ATTR_SIZE) {
2140                 dout("setattr %p size %lld -> %lld\n", inode,
2141                      inode->i_size, attr->ia_size);
2142                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2143                     attr->ia_size > inode->i_size) {
2144                         i_size_write(inode, attr->ia_size);
2145                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2146                         ci->i_reported_size = attr->ia_size;
2147                         dirtied |= CEPH_CAP_FILE_EXCL;
2148                         ia_valid |= ATTR_MTIME;
2149                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2150                            attr->ia_size != inode->i_size) {
2151                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2152                         req->r_args.setattr.old_size =
2153                                 cpu_to_le64(inode->i_size);
2154                         mask |= CEPH_SETATTR_SIZE;
2155                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2156                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2157                 }
2158         }
2159         if (ia_valid & ATTR_MTIME) {
2160                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2161                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2162                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2163                 if (issued & CEPH_CAP_FILE_EXCL) {
2164                         ci->i_time_warp_seq++;
2165                         inode->i_mtime = attr->ia_mtime;
2166                         dirtied |= CEPH_CAP_FILE_EXCL;
2167                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2168                            timespec64_compare(&inode->i_mtime,
2169                                             &attr->ia_mtime) < 0) {
2170                         inode->i_mtime = attr->ia_mtime;
2171                         dirtied |= CEPH_CAP_FILE_WR;
2172                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2173                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2174                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2175                                                &attr->ia_mtime);
2176                         mask |= CEPH_SETATTR_MTIME;
2177                         release |= CEPH_CAP_FILE_SHARED |
2178                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2179                 }
2180         }
2181
2182         /* these do nothing */
2183         if (ia_valid & ATTR_CTIME) {
2184                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2185                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2186                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2187                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2188                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2189                      only ? "ctime only" : "ignored");
2190                 if (only) {
2191                         /*
2192                          * if kernel wants to dirty ctime but nothing else,
2193                          * we need to choose a cap to dirty under, or do
2194                          * a almost-no-op setattr
2195                          */
2196                         if (issued & CEPH_CAP_AUTH_EXCL)
2197                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2198                         else if (issued & CEPH_CAP_FILE_EXCL)
2199                                 dirtied |= CEPH_CAP_FILE_EXCL;
2200                         else if (issued & CEPH_CAP_XATTR_EXCL)
2201                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2202                         else
2203                                 mask |= CEPH_SETATTR_CTIME;
2204                 }
2205         }
2206         if (ia_valid & ATTR_FILE)
2207                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2208
2209         if (dirtied) {
2210                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2211                                                            &prealloc_cf);
2212                 inode->i_ctime = attr->ia_ctime;
2213         }
2214
2215         release &= issued;
2216         spin_unlock(&ci->i_ceph_lock);
2217         if (lock_snap_rwsem)
2218                 up_read(&mdsc->snap_rwsem);
2219
2220         if (inode_dirty_flags)
2221                 __mark_inode_dirty(inode, inode_dirty_flags);
2222
2223
2224         if (mask) {
2225                 req->r_inode = inode;
2226                 ihold(inode);
2227                 req->r_inode_drop = release;
2228                 req->r_args.setattr.mask = cpu_to_le32(mask);
2229                 req->r_num_caps = 1;
2230                 req->r_stamp = attr->ia_ctime;
2231                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2232         }
2233         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2234              ceph_cap_string(dirtied), mask);
2235
2236         ceph_mdsc_put_request(req);
2237         ceph_free_cap_flush(prealloc_cf);
2238
2239         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2240                 __ceph_do_pending_vmtruncate(inode);
2241
2242         return err;
2243 }
2244
2245 /*
2246  * setattr
2247  */
2248 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2249 {
2250         struct inode *inode = d_inode(dentry);
2251         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2252         int err;
2253
2254         if (ceph_snap(inode) != CEPH_NOSNAP)
2255                 return -EROFS;
2256
2257         err = setattr_prepare(dentry, attr);
2258         if (err != 0)
2259                 return err;
2260
2261         if ((attr->ia_valid & ATTR_SIZE) &&
2262             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2263                 return -EFBIG;
2264
2265         if ((attr->ia_valid & ATTR_SIZE) &&
2266             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2267                 return -EDQUOT;
2268
2269         err = __ceph_setattr(inode, attr);
2270
2271         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2272                 err = posix_acl_chmod(inode, attr->ia_mode);
2273
2274         return err;
2275 }
2276
2277 /*
2278  * Verify that we have a lease on the given mask.  If not,
2279  * do a getattr against an mds.
2280  */
2281 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2282                       int mask, bool force)
2283 {
2284         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2285         struct ceph_mds_client *mdsc = fsc->mdsc;
2286         struct ceph_mds_request *req;
2287         int mode;
2288         int err;
2289
2290         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2291                 dout("do_getattr inode %p SNAPDIR\n", inode);
2292                 return 0;
2293         }
2294
2295         dout("do_getattr inode %p mask %s mode 0%o\n",
2296              inode, ceph_cap_string(mask), inode->i_mode);
2297         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2298                         return 0;
2299
2300         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2301         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2302         if (IS_ERR(req))
2303                 return PTR_ERR(req);
2304         req->r_inode = inode;
2305         ihold(inode);
2306         req->r_num_caps = 1;
2307         req->r_args.getattr.mask = cpu_to_le32(mask);
2308         req->r_locked_page = locked_page;
2309         err = ceph_mdsc_do_request(mdsc, NULL, req);
2310         if (locked_page && err == 0) {
2311                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2312                 if (inline_version == 0) {
2313                         /* the reply is supposed to contain inline data */
2314                         err = -EINVAL;
2315                 } else if (inline_version == CEPH_INLINE_NONE) {
2316                         err = -ENODATA;
2317                 } else {
2318                         err = req->r_reply_info.targeti.inline_len;
2319                 }
2320         }
2321         ceph_mdsc_put_request(req);
2322         dout("do_getattr result=%d\n", err);
2323         return err;
2324 }
2325
2326
2327 /*
2328  * Check inode permissions.  We verify we have a valid value for
2329  * the AUTH cap, then call the generic handler.
2330  */
2331 int ceph_permission(struct inode *inode, int mask)
2332 {
2333         int err;
2334
2335         if (mask & MAY_NOT_BLOCK)
2336                 return -ECHILD;
2337
2338         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2339
2340         if (!err)
2341                 err = generic_permission(inode, mask);
2342         return err;
2343 }
2344
2345 /* Craft a mask of needed caps given a set of requested statx attrs. */
2346 static int statx_to_caps(u32 want)
2347 {
2348         int mask = 0;
2349
2350         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2351                 mask |= CEPH_CAP_AUTH_SHARED;
2352
2353         if (want & (STATX_NLINK|STATX_CTIME))
2354                 mask |= CEPH_CAP_LINK_SHARED;
2355
2356         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2357                     STATX_BLOCKS))
2358                 mask |= CEPH_CAP_FILE_SHARED;
2359
2360         if (want & (STATX_CTIME))
2361                 mask |= CEPH_CAP_XATTR_SHARED;
2362
2363         return mask;
2364 }
2365
2366 /*
2367  * Get all the attributes. If we have sufficient caps for the requested attrs,
2368  * then we can avoid talking to the MDS at all.
2369  */
2370 int ceph_getattr(const struct path *path, struct kstat *stat,
2371                  u32 request_mask, unsigned int flags)
2372 {
2373         struct inode *inode = d_inode(path->dentry);
2374         struct ceph_inode_info *ci = ceph_inode(inode);
2375         u32 valid_mask = STATX_BASIC_STATS;
2376         int err = 0;
2377
2378         /* Skip the getattr altogether if we're asked not to sync */
2379         if (!(flags & AT_STATX_DONT_SYNC)) {
2380                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2381                                       flags & AT_STATX_FORCE_SYNC);
2382                 if (err)
2383                         return err;
2384         }
2385
2386         generic_fillattr(inode, stat);
2387         stat->ino = ceph_present_inode(inode);
2388
2389         /*
2390          * btime on newly-allocated inodes is 0, so if this is still set to
2391          * that, then assume that it's not valid.
2392          */
2393         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2394                 stat->btime = ci->i_btime;
2395                 valid_mask |= STATX_BTIME;
2396         }
2397
2398         if (ceph_snap(inode) == CEPH_NOSNAP)
2399                 stat->dev = inode->i_sb->s_dev;
2400         else
2401                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2402
2403         if (S_ISDIR(inode->i_mode)) {
2404                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2405                                         RBYTES))
2406                         stat->size = ci->i_rbytes;
2407                 else
2408                         stat->size = ci->i_files + ci->i_subdirs;
2409                 stat->blocks = 0;
2410                 stat->blksize = 65536;
2411                 /*
2412                  * Some applications rely on the number of st_nlink
2413                  * value on directories to be either 0 (if unlinked)
2414                  * or 2 + number of subdirectories.
2415                  */
2416                 if (stat->nlink == 1)
2417                         /* '.' + '..' + subdirs */
2418                         stat->nlink = 1 + 1 + ci->i_subdirs;
2419         }
2420
2421         stat->result_mask = request_mask & valid_mask;
2422         return err;
2423 }