fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16 #include <linux/iversion.h>
  17
  18 #include "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21 #include <linux/ceph/decode.h>
  22
  23 /*
  24  * Ceph inode operations
  25  *
  26  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  27  * setattr, etc.), xattr helpers, and helpers for assimilating
  28  * metadata returned by the MDS into our cache.
  29  *
  30  * Also define helpers for doing asynchronous writeback, invalidation,
  31  * and truncation for the benefit of those who can't afford to block
  32  * (typically because they are in the message handler path).
  33  */
  34
  35 static const struct inode_operations ceph_symlink_iops;
  36
  37 static void ceph_inode_work(struct work_struct *work);
  38
  39 /*
  40  * find or create an inode, given the ceph ino number
  41  */
  42 static int ceph_set_ino_cb(struct inode *inode, void *data)
  43 {
  44         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  45         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  46         inode_set_iversion_raw(inode, 0);
  47         return 0;
  48 }
  49
  50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  51 {
  52         struct inode *inode;
  53         ino_t t = ceph_vino_to_ino(vino);
  54
  55         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  56         if (!inode)
  57                 return ERR_PTR(-ENOMEM);
  58         if (inode->i_state & I_NEW)
  59                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  60                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  61
  62         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  63              vino.snap, inode);
  64         return inode;
  65 }
  66
  67 /*
  68  * get/constuct snapdir inode for a given directory
  69  */
  70 struct inode *ceph_get_snapdir(struct inode *parent)
  71 {
  72         struct ceph_vino vino = {
  73                 .ino = ceph_ino(parent),
  74                 .snap = CEPH_SNAPDIR,
  75         };
  76         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  77         struct ceph_inode_info *ci = ceph_inode(inode);
  78
  79         BUG_ON(!S_ISDIR(parent->i_mode));
  80         if (IS_ERR(inode))
  81                 return inode;
  82         inode->i_mode = parent->i_mode;
  83         inode->i_uid = parent->i_uid;
  84         inode->i_gid = parent->i_gid;
  85         inode->i_mtime = parent->i_mtime;
  86         inode->i_ctime = parent->i_ctime;
  87         inode->i_atime = parent->i_atime;
  88         inode->i_op = &ceph_snapdir_iops;
  89         inode->i_fop = &ceph_snapdir_fops;
  90         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  91         ci->i_rbytes = 0;
  92         ci->i_btime = ceph_inode(parent)->i_btime;
  93
  94         if (inode->i_state & I_NEW)
  95                 unlock_new_inode(inode);
  96
  97         return inode;
  98 }
  99
 100 const struct inode_operations ceph_file_iops = {
 101         .permission = ceph_permission,
 102         .setattr = ceph_setattr,
 103         .getattr = ceph_getattr,
 104         .listxattr = ceph_listxattr,
 105         .get_acl = ceph_get_acl,
 106         .set_acl = ceph_set_acl,
 107 };
 108
 109
 110 /*
 111  * We use a 'frag tree' to keep track of the MDS's directory fragments
 112  * for a given inode (usually there is just a single fragment).  We
 113  * need to know when a child frag is delegated to a new MDS, or when
 114  * it is flagged as replicated, so we can direct our requests
 115  * accordingly.
 116  */
 117
 118 /*
 119  * find/create a frag in the tree
 120  */
 121 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 122                                                     u32 f)
 123 {
 124         struct rb_node **p;
 125         struct rb_node *parent = NULL;
 126         struct ceph_inode_frag *frag;
 127         int c;
 128
 129         p = &ci->i_fragtree.rb_node;
 130         while (*p) {
 131                 parent = *p;
 132                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 133                 c = ceph_frag_compare(f, frag->frag);
 134                 if (c < 0)
 135                         p = &(*p)->rb_left;
 136                 else if (c > 0)
 137                         p = &(*p)->rb_right;
 138                 else
 139                         return frag;
 140         }
 141
 142         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 143         if (!frag)
 144                 return ERR_PTR(-ENOMEM);
 145
 146         frag->frag = f;
 147         frag->split_by = 0;
 148         frag->mds = -1;
 149         frag->ndist = 0;
 150
 151         rb_link_node(&frag->node, parent, p);
 152         rb_insert_color(&frag->node, &ci->i_fragtree);
 153
 154         dout("get_or_create_frag added %llx.%llx frag %x\n",
 155              ceph_vinop(&ci->vfs_inode), f);
 156         return frag;
 157 }
 158
 159 /*
 160  * find a specific frag @f
 161  */
 162 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 163 {
 164         struct rb_node *n = ci->i_fragtree.rb_node;
 165
 166         while (n) {
 167                 struct ceph_inode_frag *frag =
 168                         rb_entry(n, struct ceph_inode_frag, node);
 169                 int c = ceph_frag_compare(f, frag->frag);
 170                 if (c < 0)
 171                         n = n->rb_left;
 172                 else if (c > 0)
 173                         n = n->rb_right;
 174                 else
 175                         return frag;
 176         }
 177         return NULL;
 178 }
 179
 180 /*
 181  * Choose frag containing the given value @v.  If @pfrag is
 182  * specified, copy the frag delegation info to the caller if
 183  * it is present.
 184  */
 185 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 186                               struct ceph_inode_frag *pfrag, int *found)
 187 {
 188         u32 t = ceph_frag_make(0, 0);
 189         struct ceph_inode_frag *frag;
 190         unsigned nway, i;
 191         u32 n;
 192
 193         if (found)
 194                 *found = 0;
 195
 196         while (1) {
 197                 WARN_ON(!ceph_frag_contains_value(t, v));
 198                 frag = __ceph_find_frag(ci, t);
 199                 if (!frag)
 200                         break; /* t is a leaf */
 201                 if (frag->split_by == 0) {
 202                         if (pfrag)
 203                                 memcpy(pfrag, frag, sizeof(*pfrag));
 204                         if (found)
 205                                 *found = 1;
 206                         break;
 207                 }
 208
 209                 /* choose child */
 210                 nway = 1 << frag->split_by;
 211                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 212                      frag->split_by, nway);
 213                 for (i = 0; i < nway; i++) {
 214                         n = ceph_frag_make_child(t, frag->split_by, i);
 215                         if (ceph_frag_contains_value(n, v)) {
 216                                 t = n;
 217                                 break;
 218                         }
 219                 }
 220                 BUG_ON(i == nway);
 221         }
 222         dout("choose_frag(%x) = %x\n", v, t);
 223
 224         return t;
 225 }
 226
 227 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 228                      struct ceph_inode_frag *pfrag, int *found)
 229 {
 230         u32 ret;
 231         mutex_lock(&ci->i_fragtree_mutex);
 232         ret = __ceph_choose_frag(ci, v, pfrag, found);
 233         mutex_unlock(&ci->i_fragtree_mutex);
 234         return ret;
 235 }
 236
 237 /*
 238  * Process dirfrag (delegation) info from the mds.  Include leaf
 239  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 240  * branches/splits are included in i_fragtree)
 241  */
 242 static int ceph_fill_dirfrag(struct inode *inode,
 243                              struct ceph_mds_reply_dirfrag *dirinfo)
 244 {
 245         struct ceph_inode_info *ci = ceph_inode(inode);
 246         struct ceph_inode_frag *frag;
 247         u32 id = le32_to_cpu(dirinfo->frag);
 248         int mds = le32_to_cpu(dirinfo->auth);
 249         int ndist = le32_to_cpu(dirinfo->ndist);
 250         int diri_auth = -1;
 251         int i;
 252         int err = 0;
 253
 254         spin_lock(&ci->i_ceph_lock);
 255         if (ci->i_auth_cap)
 256                 diri_auth = ci->i_auth_cap->mds;
 257         spin_unlock(&ci->i_ceph_lock);
 258
 259         if (mds == -1) /* CDIR_AUTH_PARENT */
 260                 mds = diri_auth;
 261
 262         mutex_lock(&ci->i_fragtree_mutex);
 263         if (ndist == 0 && mds == diri_auth) {
 264                 /* no delegation info needed. */
 265                 frag = __ceph_find_frag(ci, id);
 266                 if (!frag)
 267                         goto out;
 268                 if (frag->split_by == 0) {
 269                         /* tree leaf, remove */
 270                         dout("fill_dirfrag removed %llx.%llx frag %x"
 271                              " (no ref)\n", ceph_vinop(inode), id);
 272                         rb_erase(&frag->node, &ci->i_fragtree);
 273                         kfree(frag);
 274                 } else {
 275                         /* tree branch, keep and clear */
 276                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 277                              " referral\n", ceph_vinop(inode), id);
 278                         frag->mds = -1;
 279                         frag->ndist = 0;
 280                 }
 281                 goto out;
 282         }
 283
 284
 285         /* find/add this frag to store mds delegation info */
 286         frag = __get_or_create_frag(ci, id);
 287         if (IS_ERR(frag)) {
 288                 /* this is not the end of the world; we can continue
 289                    with bad/inaccurate delegation info */
 290                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 291                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 292                 err = -ENOMEM;
 293                 goto out;
 294         }
 295
 296         frag->mds = mds;
 297         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 298         for (i = 0; i < frag->ndist; i++)
 299                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 300         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 301              ceph_vinop(inode), frag->frag, frag->ndist);
 302
 303 out:
 304         mutex_unlock(&ci->i_fragtree_mutex);
 305         return err;
 306 }
 307
 308 static int frag_tree_split_cmp(const void *l, const void *r)
 309 {
 310         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 311         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 312         return ceph_frag_compare(le32_to_cpu(ls->frag),
 313                                  le32_to_cpu(rs->frag));
 314 }
 315
 316 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 317 {
 318         if (!frag)
 319                 return f == ceph_frag_make(0, 0);
 320         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 321                 return false;
 322         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 323 }
 324
 325 static int ceph_fill_fragtree(struct inode *inode,
 326                               struct ceph_frag_tree_head *fragtree,
 327                               struct ceph_mds_reply_dirfrag *dirinfo)
 328 {
 329         struct ceph_inode_info *ci = ceph_inode(inode);
 330         struct ceph_inode_frag *frag, *prev_frag = NULL;
 331         struct rb_node *rb_node;
 332         unsigned i, split_by, nsplits;
 333         u32 id;
 334         bool update = false;
 335
 336         mutex_lock(&ci->i_fragtree_mutex);
 337         nsplits = le32_to_cpu(fragtree->nsplits);
 338         if (nsplits != ci->i_fragtree_nsplits) {
 339                 update = true;
 340         } else if (nsplits) {
 341                 i = prandom_u32() % nsplits;
 342                 id = le32_to_cpu(fragtree->splits[i].frag);
 343                 if (!__ceph_find_frag(ci, id))
 344                         update = true;
 345         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 346                 rb_node = rb_first(&ci->i_fragtree);
 347                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 348                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 349                         update = true;
 350         }
 351         if (!update && dirinfo) {
 352                 id = le32_to_cpu(dirinfo->frag);
 353                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 354                         update = true;
 355         }
 356         if (!update)
 357                 goto out_unlock;
 358
 359         if (nsplits > 1) {
 360                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 361                      frag_tree_split_cmp, NULL);
 362         }
 363
 364         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 365         rb_node = rb_first(&ci->i_fragtree);
 366         for (i = 0; i < nsplits; i++) {
 367                 id = le32_to_cpu(fragtree->splits[i].frag);
 368                 split_by = le32_to_cpu(fragtree->splits[i].by);
 369                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 370                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 371                                "frag %x split by %d\n", ceph_vinop(inode),
 372                                i, nsplits, id, split_by);
 373                         continue;
 374                 }
 375                 frag = NULL;
 376                 while (rb_node) {
 377                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 378                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 379                                 if (frag->frag != id)
 380                                         frag = NULL;
 381                                 else
 382                                         rb_node = rb_next(rb_node);
 383                                 break;
 384                         }
 385                         rb_node = rb_next(rb_node);
 386                         /* delete stale split/leaf node */
 387                         if (frag->split_by > 0 ||
 388                             !is_frag_child(frag->frag, prev_frag)) {
 389                                 rb_erase(&frag->node, &ci->i_fragtree);
 390                                 if (frag->split_by > 0)
 391                                         ci->i_fragtree_nsplits--;
 392                                 kfree(frag);
 393                         }
 394                         frag = NULL;
 395                 }
 396                 if (!frag) {
 397                         frag = __get_or_create_frag(ci, id);
 398                         if (IS_ERR(frag))
 399                                 continue;
 400                 }
 401                 if (frag->split_by == 0)
 402                         ci->i_fragtree_nsplits++;
 403                 frag->split_by = split_by;
 404                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 405                 prev_frag = frag;
 406         }
 407         while (rb_node) {
 408                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 409                 rb_node = rb_next(rb_node);
 410                 /* delete stale split/leaf node */
 411                 if (frag->split_by > 0 ||
 412                     !is_frag_child(frag->frag, prev_frag)) {
 413                         rb_erase(&frag->node, &ci->i_fragtree);
 414                         if (frag->split_by > 0)
 415                                 ci->i_fragtree_nsplits--;
 416                         kfree(frag);
 417                 }
 418         }
 419 out_unlock:
 420         mutex_unlock(&ci->i_fragtree_mutex);
 421         return 0;
 422 }
 423
 424 /*
 425  * initialize a newly allocated inode.
 426  */
 427 struct inode *ceph_alloc_inode(struct super_block *sb)
 428 {
 429         struct ceph_inode_info *ci;
 430         int i;
 431
 432         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 433         if (!ci)
 434                 return NULL;
 435
 436         dout("alloc_inode %p\n", &ci->vfs_inode);
 437
 438         spin_lock_init(&ci->i_ceph_lock);
 439
 440         ci->i_version = 0;
 441         ci->i_inline_version = 0;
 442         ci->i_time_warp_seq = 0;
 443         ci->i_ceph_flags = 0;
 444         atomic64_set(&ci->i_ordered_count, 1);
 445         atomic64_set(&ci->i_release_count, 1);
 446         atomic64_set(&ci->i_complete_seq[0], 0);
 447         atomic64_set(&ci->i_complete_seq[1], 0);
 448         ci->i_symlink = NULL;
 449
 450         ci->i_max_bytes = 0;
 451         ci->i_max_files = 0;
 452
 453         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 454         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 455         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 456
 457         ci->i_fragtree = RB_ROOT;
 458         mutex_init(&ci->i_fragtree_mutex);
 459
 460         ci->i_xattrs.blob = NULL;
 461         ci->i_xattrs.prealloc_blob = NULL;
 462         ci->i_xattrs.dirty = false;
 463         ci->i_xattrs.index = RB_ROOT;
 464         ci->i_xattrs.count = 0;
 465         ci->i_xattrs.names_size = 0;
 466         ci->i_xattrs.vals_size = 0;
 467         ci->i_xattrs.version = 0;
 468         ci->i_xattrs.index_version = 0;
 469
 470         ci->i_caps = RB_ROOT;
 471         ci->i_auth_cap = NULL;
 472         ci->i_dirty_caps = 0;
 473         ci->i_flushing_caps = 0;
 474         INIT_LIST_HEAD(&ci->i_dirty_item);
 475         INIT_LIST_HEAD(&ci->i_flushing_item);
 476         ci->i_prealloc_cap_flush = NULL;
 477         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 478         init_waitqueue_head(&ci->i_cap_wq);
 479         ci->i_hold_caps_max = 0;
 480         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 481         INIT_LIST_HEAD(&ci->i_cap_snaps);
 482         ci->i_head_snapc = NULL;
 483         ci->i_snap_caps = 0;
 484
 485         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 486         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 487                 ci->i_nr_by_mode[i] = 0;
 488
 489         mutex_init(&ci->i_truncate_mutex);
 490         ci->i_truncate_seq = 0;
 491         ci->i_truncate_size = 0;
 492         ci->i_truncate_pending = 0;
 493
 494         ci->i_max_size = 0;
 495         ci->i_reported_size = 0;
 496         ci->i_wanted_max_size = 0;
 497         ci->i_requested_max_size = 0;
 498
 499         ci->i_pin_ref = 0;
 500         ci->i_rd_ref = 0;
 501         ci->i_rdcache_ref = 0;
 502         ci->i_wr_ref = 0;
 503         ci->i_wb_ref = 0;
 504         ci->i_fx_ref = 0;
 505         ci->i_wrbuffer_ref = 0;
 506         ci->i_wrbuffer_ref_head = 0;
 507         atomic_set(&ci->i_filelock_ref, 0);
 508         atomic_set(&ci->i_shared_gen, 1);
 509         ci->i_rdcache_gen = 0;
 510         ci->i_rdcache_revoking = 0;
 511
 512         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 513         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 514         spin_lock_init(&ci->i_unsafe_lock);
 515
 516         ci->i_snap_realm = NULL;
 517         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 518         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 519
 520         INIT_WORK(&ci->i_work, ceph_inode_work);
 521         ci->i_work_mask = 0;
 522         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 523
 524         ceph_fscache_inode_init(ci);
 525
 526         ci->i_meta_err = 0;
 527
 528         return &ci->vfs_inode;
 529 }
 530
 531 void ceph_free_inode(struct inode *inode)
 532 {
 533         struct ceph_inode_info *ci = ceph_inode(inode);
 534
 535         kfree(ci->i_symlink);
 536         kmem_cache_free(ceph_inode_cachep, ci);
 537 }
 538
 539 void ceph_evict_inode(struct inode *inode)
 540 {
 541         struct ceph_inode_info *ci = ceph_inode(inode);
 542         struct ceph_inode_frag *frag;
 543         struct rb_node *n;
 544
 545         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 546
 547         truncate_inode_pages_final(&inode->i_data);
 548         clear_inode(inode);
 549
 550         ceph_fscache_unregister_inode_cookie(ci);
 551
 552         __ceph_remove_caps(ci);
 553
 554         if (__ceph_has_any_quota(ci))
 555                 ceph_adjust_quota_realms_count(inode, false);
 556
 557         /*
 558          * we may still have a snap_realm reference if there are stray
 559          * caps in i_snap_caps.
 560          */
 561         if (ci->i_snap_realm) {
 562                 struct ceph_mds_client *mdsc =
 563                                         ceph_inode_to_client(inode)->mdsc;
 564                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 565                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 566                         dout(" dropping residual ref to snap realm %p\n",
 567                              realm);
 568                         spin_lock(&realm->inodes_with_caps_lock);
 569                         list_del_init(&ci->i_snap_realm_item);
 570                         ci->i_snap_realm = NULL;
 571                         if (realm->ino == ci->i_vino.ino)
 572                                 realm->inode = NULL;
 573                         spin_unlock(&realm->inodes_with_caps_lock);
 574                         ceph_put_snap_realm(mdsc, realm);
 575                 } else {
 576                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 577                         ci->i_snap_realm = NULL;
 578                 }
 579         }
 580
 581         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 582                 frag = rb_entry(n, struct ceph_inode_frag, node);
 583                 rb_erase(n, &ci->i_fragtree);
 584                 kfree(frag);
 585         }
 586         ci->i_fragtree_nsplits = 0;
 587
 588         __ceph_destroy_xattrs(ci);
 589         if (ci->i_xattrs.blob)
 590                 ceph_buffer_put(ci->i_xattrs.blob);
 591         if (ci->i_xattrs.prealloc_blob)
 592                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 593
 594         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 595         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 596 }
 597
 598 static inline blkcnt_t calc_inode_blocks(u64 size)
 599 {
 600         return (size + (1<<9) - 1) >> 9;
 601 }
 602
 603 /*
 604  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 605  * careful because either the client or MDS may have more up to date
 606  * info, depending on which capabilities are held, and whether
 607  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 608  * and size are monotonically increasing, except when utimes() or
 609  * truncate() increments the corresponding _seq values.)
 610  */
 611 int ceph_fill_file_size(struct inode *inode, int issued,
 612                         u32 truncate_seq, u64 truncate_size, u64 size)
 613 {
 614         struct ceph_inode_info *ci = ceph_inode(inode);
 615         int queue_trunc = 0;
 616
 617         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 618             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 619                 dout("size %lld -> %llu\n", inode->i_size, size);
 620                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 621                         pr_err("fill_file_size non-zero size for directory\n");
 622                         size = 0;
 623                 }
 624                 i_size_write(inode, size);
 625                 inode->i_blocks = calc_inode_blocks(size);
 626                 ci->i_reported_size = size;
 627                 if (truncate_seq != ci->i_truncate_seq) {
 628                         dout("truncate_seq %u -> %u\n",
 629                              ci->i_truncate_seq, truncate_seq);
 630                         ci->i_truncate_seq = truncate_seq;
 631
 632                         /* the MDS should have revoked these caps */
 633                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 634                                                CEPH_CAP_FILE_RD |
 635                                                CEPH_CAP_FILE_WR |
 636                                                CEPH_CAP_FILE_LAZYIO));
 637                         /*
 638                          * If we hold relevant caps, or in the case where we're
 639                          * not the only client referencing this file and we
 640                          * don't hold those caps, then we need to check whether
 641                          * the file is either opened or mmaped
 642                          */
 643                         if ((issued & (CEPH_CAP_FILE_CACHE|
 644                                        CEPH_CAP_FILE_BUFFER)) ||
 645                             mapping_mapped(inode->i_mapping) ||
 646                             __ceph_is_file_opened(ci)) {
 647                                 ci->i_truncate_pending++;
 648                                 queue_trunc = 1;
 649                         }
 650                 }
 651         }
 652         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 653             ci->i_truncate_size != truncate_size) {
 654                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 655                      truncate_size);
 656                 ci->i_truncate_size = truncate_size;
 657         }
 658
 659         if (queue_trunc)
 660                 ceph_fscache_invalidate(inode);
 661
 662         return queue_trunc;
 663 }
 664
 665 void ceph_fill_file_time(struct inode *inode, int issued,
 666                          u64 time_warp_seq, struct timespec64 *ctime,
 667                          struct timespec64 *mtime, struct timespec64 *atime)
 668 {
 669         struct ceph_inode_info *ci = ceph_inode(inode);
 670         int warn = 0;
 671
 672         if (issued & (CEPH_CAP_FILE_EXCL|
 673                       CEPH_CAP_FILE_WR|
 674                       CEPH_CAP_FILE_BUFFER|
 675                       CEPH_CAP_AUTH_EXCL|
 676                       CEPH_CAP_XATTR_EXCL)) {
 677                 if (ci->i_version == 0 ||
 678                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 679                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 680                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 681                              ctime->tv_sec, ctime->tv_nsec);
 682                         inode->i_ctime = *ctime;
 683                 }
 684                 if (ci->i_version == 0 ||
 685                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 686                         /* the MDS did a utimes() */
 687                         dout("mtime %lld.%09ld -> %lld.%09ld "
 688                              "tw %d -> %d\n",
 689                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 690                              mtime->tv_sec, mtime->tv_nsec,
 691                              ci->i_time_warp_seq, (int)time_warp_seq);
 692
 693                         inode->i_mtime = *mtime;
 694                         inode->i_atime = *atime;
 695                         ci->i_time_warp_seq = time_warp_seq;
 696                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 697                         /* nobody did utimes(); take the max */
 698                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 699                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 700                                      inode->i_mtime.tv_sec,
 701                                      inode->i_mtime.tv_nsec,
 702                                      mtime->tv_sec, mtime->tv_nsec);
 703                                 inode->i_mtime = *mtime;
 704                         }
 705                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 706                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 707                                      inode->i_atime.tv_sec,
 708                                      inode->i_atime.tv_nsec,
 709                                      atime->tv_sec, atime->tv_nsec);
 710                                 inode->i_atime = *atime;
 711                         }
 712                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 713                         /* we did a utimes(); ignore mds values */
 714                 } else {
 715                         warn = 1;
 716                 }
 717         } else {
 718                 /* we have no write|excl caps; whatever the MDS says is true */
 719                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 720                         inode->i_ctime = *ctime;
 721                         inode->i_mtime = *mtime;
 722                         inode->i_atime = *atime;
 723                         ci->i_time_warp_seq = time_warp_seq;
 724                 } else {
 725                         warn = 1;
 726                 }
 727         }
 728         if (warn) /* time_warp_seq shouldn't go backwards */
 729                 dout("%p mds time_warp_seq %llu < %u\n",
 730                      inode, time_warp_seq, ci->i_time_warp_seq);
 731 }
 732
 733 /*
 734  * Populate an inode based on info from mds.  May be called on new or
 735  * existing inodes.
 736  */
 737 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 738                     struct ceph_mds_reply_info_in *iinfo,
 739                     struct ceph_mds_reply_dirfrag *dirinfo,
 740                     struct ceph_mds_session *session, int cap_fmode,
 741                     struct ceph_cap_reservation *caps_reservation)
 742 {
 743         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 744         struct ceph_mds_reply_inode *info = iinfo->in;
 745         struct ceph_inode_info *ci = ceph_inode(inode);
 746         int issued, new_issued, info_caps;
 747         struct timespec64 mtime, atime, ctime;
 748         struct ceph_buffer *xattr_blob = NULL;
 749         struct ceph_buffer *old_blob = NULL;
 750         struct ceph_string *pool_ns = NULL;
 751         struct ceph_cap *new_cap = NULL;
 752         int err = 0;
 753         bool wake = false;
 754         bool queue_trunc = false;
 755         bool new_version = false;
 756         bool fill_inline = false;
 757
 758         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 759              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 760              ci->i_version);
 761
 762         info_caps = le32_to_cpu(info->cap.caps);
 763
 764         /* prealloc new cap struct */
 765         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 766                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 767                 if (!new_cap)
 768                         return -ENOMEM;
 769         }
 770
 771         /*
 772          * prealloc xattr data, if it looks like we'll need it.  only
 773          * if len > 4 (meaning there are actually xattrs; the first 4
 774          * bytes are the xattr count).
 775          */
 776         if (iinfo->xattr_len > 4) {
 777                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 778                 if (!xattr_blob)
 779                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 780                                iinfo->xattr_len);
 781         }
 782
 783         if (iinfo->pool_ns_len > 0)
 784                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 785                                                      iinfo->pool_ns_len);
 786
 787         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 788                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 789
 790         spin_lock(&ci->i_ceph_lock);
 791
 792         /*
 793          * provided version will be odd if inode value is projected,
 794          * even if stable.  skip the update if we have newer stable
 795          * info (ours>=theirs, e.g. due to racing mds replies), unless
 796          * we are getting projected (unstable) info (in which case the
 797          * version is odd, and we want ours>theirs).
 798          *   us   them
 799          *   2    2     skip
 800          *   3    2     skip
 801          *   3    3     update
 802          */
 803         if (ci->i_version == 0 ||
 804             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 805              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 806                 new_version = true;
 807
 808         /* Update change_attribute */
 809         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 810
 811         __ceph_caps_issued(ci, &issued);
 812         issued |= __ceph_caps_dirty(ci);
 813         new_issued = ~issued & info_caps;
 814
 815         /* update inode */
 816         inode->i_rdev = le32_to_cpu(info->rdev);
 817         /* directories have fl_stripe_unit set to zero */
 818         if (le32_to_cpu(info->layout.fl_stripe_unit))
 819                 inode->i_blkbits =
 820                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 821         else
 822                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 823
 824         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 825
 826         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 827             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 828                 inode->i_mode = le32_to_cpu(info->mode);
 829                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 830                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 831                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 832                      from_kuid(&init_user_ns, inode->i_uid),
 833                      from_kgid(&init_user_ns, inode->i_gid));
 834                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 835                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 836         }
 837
 838         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 839             (issued & CEPH_CAP_LINK_EXCL) == 0)
 840                 set_nlink(inode, le32_to_cpu(info->nlink));
 841
 842         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 843                 /* be careful with mtime, atime, size */
 844                 ceph_decode_timespec64(&atime, &info->atime);
 845                 ceph_decode_timespec64(&mtime, &info->mtime);
 846                 ceph_decode_timespec64(&ctime, &info->ctime);
 847                 ceph_fill_file_time(inode, issued,
 848                                 le32_to_cpu(info->time_warp_seq),
 849                                 &ctime, &mtime, &atime);
 850         }
 851
 852         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 853                 ci->i_files = le64_to_cpu(info->files);
 854                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 855         }
 856
 857         if (new_version ||
 858             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 859                 s64 old_pool = ci->i_layout.pool_id;
 860                 struct ceph_string *old_ns;
 861
 862                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 863                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 864                                         lockdep_is_held(&ci->i_ceph_lock));
 865                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 866
 867                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 868                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 869
 870                 pool_ns = old_ns;
 871
 872                 queue_trunc = ceph_fill_file_size(inode, issued,
 873                                         le32_to_cpu(info->truncate_seq),
 874                                         le64_to_cpu(info->truncate_size),
 875                                         le64_to_cpu(info->size));
 876                 /* only update max_size on auth cap */
 877                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 878                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 879                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 880                                         le64_to_cpu(info->max_size));
 881                         ci->i_max_size = le64_to_cpu(info->max_size);
 882                 }
 883         }
 884
 885         /* layout and rstat are not tracked by capability, update them if
 886          * the inode info is from auth mds */
 887         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 888                 if (S_ISDIR(inode->i_mode)) {
 889                         ci->i_dir_layout = iinfo->dir_layout;
 890                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 891                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 892                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 893                         ci->i_dir_pin = iinfo->dir_pin;
 894                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 895                 }
 896         }
 897
 898         /* xattrs */
 899         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 900         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 901             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 902                 if (ci->i_xattrs.blob)
 903                         old_blob = ci->i_xattrs.blob;
 904                 ci->i_xattrs.blob = xattr_blob;
 905                 if (xattr_blob)
 906                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 907                                iinfo->xattr_data, iinfo->xattr_len);
 908                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 909                 ceph_forget_all_cached_acls(inode);
 910                 ceph_security_invalidate_secctx(inode);
 911                 xattr_blob = NULL;
 912         }
 913
 914         /* finally update i_version */
 915         if (le64_to_cpu(info->version) > ci->i_version)
 916                 ci->i_version = le64_to_cpu(info->version);
 917
 918         inode->i_mapping->a_ops = &ceph_aops;
 919
 920         switch (inode->i_mode & S_IFMT) {
 921         case S_IFIFO:
 922         case S_IFBLK:
 923         case S_IFCHR:
 924         case S_IFSOCK:
 925                 inode->i_blkbits = PAGE_SHIFT;
 926                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 927                 inode->i_op = &ceph_file_iops;
 928                 break;
 929         case S_IFREG:
 930                 inode->i_op = &ceph_file_iops;
 931                 inode->i_fop = &ceph_file_fops;
 932                 break;
 933         case S_IFLNK:
 934                 inode->i_op = &ceph_symlink_iops;
 935                 if (!ci->i_symlink) {
 936                         u32 symlen = iinfo->symlink_len;
 937                         char *sym;
 938
 939                         spin_unlock(&ci->i_ceph_lock);
 940
 941                         if (symlen != i_size_read(inode)) {
 942                                 pr_err("%s %llx.%llx BAD symlink "
 943                                         "size %lld\n", __func__,
 944                                         ceph_vinop(inode),
 945                                         i_size_read(inode));
 946                                 i_size_write(inode, symlen);
 947                                 inode->i_blocks = calc_inode_blocks(symlen);
 948                         }
 949
 950                         err = -ENOMEM;
 951                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 952                         if (!sym)
 953                                 goto out;
 954
 955                         spin_lock(&ci->i_ceph_lock);
 956                         if (!ci->i_symlink)
 957                                 ci->i_symlink = sym;
 958                         else
 959                                 kfree(sym); /* lost a race */
 960                 }
 961                 inode->i_link = ci->i_symlink;
 962                 break;
 963         case S_IFDIR:
 964                 inode->i_op = &ceph_dir_iops;
 965                 inode->i_fop = &ceph_dir_fops;
 966                 break;
 967         default:
 968                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
 969                        ceph_vinop(inode), inode->i_mode);
 970         }
 971
 972         /* were we issued a capability? */
 973         if (info_caps) {
 974                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 975                         ceph_add_cap(inode, session,
 976                                      le64_to_cpu(info->cap.cap_id),
 977                                      info_caps,
 978                                      le32_to_cpu(info->cap.wanted),
 979                                      le32_to_cpu(info->cap.seq),
 980                                      le32_to_cpu(info->cap.mseq),
 981                                      le64_to_cpu(info->cap.realm),
 982                                      info->cap.flags, &new_cap);
 983
 984                         /* set dir completion flag? */
 985                         if (S_ISDIR(inode->i_mode) &&
 986                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 987                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 988                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 989                             !__ceph_dir_is_complete(ci)) {
 990                                 dout(" marking %p complete (empty)\n", inode);
 991                                 i_size_write(inode, 0);
 992                                 __ceph_dir_set_complete(ci,
 993                                         atomic64_read(&ci->i_release_count),
 994                                         atomic64_read(&ci->i_ordered_count));
 995                         }
 996
 997                         wake = true;
 998                 } else {
 999                         dout(" %p got snap_caps %s\n", inode,
1000                              ceph_cap_string(info_caps));
1001                         ci->i_snap_caps |= info_caps;
1002                 }
1003         }
1004
1005         if (iinfo->inline_version > 0 &&
1006             iinfo->inline_version >= ci->i_inline_version) {
1007                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1008                 ci->i_inline_version = iinfo->inline_version;
1009                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1010                     (locked_page || (info_caps & cache_caps)))
1011                         fill_inline = true;
1012         }
1013
1014         if (cap_fmode >= 0) {
1015                 if (!info_caps)
1016                         pr_warn("mds issued no caps on %llx.%llx\n",
1017                                 ceph_vinop(inode));
1018                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1019         }
1020
1021         spin_unlock(&ci->i_ceph_lock);
1022
1023         if (fill_inline)
1024                 ceph_fill_inline_data(inode, locked_page,
1025                                       iinfo->inline_data, iinfo->inline_len);
1026
1027         if (wake)
1028                 wake_up_all(&ci->i_cap_wq);
1029
1030         /* queue truncate if we saw i_size decrease */
1031         if (queue_trunc)
1032                 ceph_queue_vmtruncate(inode);
1033
1034         /* populate frag tree */
1035         if (S_ISDIR(inode->i_mode))
1036                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1037
1038         /* update delegation info? */
1039         if (dirinfo)
1040                 ceph_fill_dirfrag(inode, dirinfo);
1041
1042         err = 0;
1043 out:
1044         if (new_cap)
1045                 ceph_put_cap(mdsc, new_cap);
1046         ceph_buffer_put(old_blob);
1047         ceph_buffer_put(xattr_blob);
1048         ceph_put_string(pool_ns);
1049         return err;
1050 }
1051
1052 /*
1053  * caller should hold session s_mutex and dentry->d_lock.
1054  */
1055 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1056                                   struct ceph_mds_reply_lease *lease,
1057                                   struct ceph_mds_session *session,
1058                                   unsigned long from_time,
1059                                   struct ceph_mds_session **old_lease_session)
1060 {
1061         struct ceph_dentry_info *di = ceph_dentry(dentry);
1062         unsigned mask = le16_to_cpu(lease->mask);
1063         long unsigned duration = le32_to_cpu(lease->duration_ms);
1064         long unsigned ttl = from_time + (duration * HZ) / 1000;
1065         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1066
1067         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1068              dentry, duration, ttl);
1069
1070         /* only track leases on regular dentries */
1071         if (ceph_snap(dir) != CEPH_NOSNAP)
1072                 return;
1073
1074         if (mask & CEPH_LEASE_PRIMARY_LINK)
1075                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1076         else
1077                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1078
1079         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1080         if (!(mask & CEPH_LEASE_VALID)) {
1081                 __ceph_dentry_dir_lease_touch(di);
1082                 return;
1083         }
1084
1085         if (di->lease_gen == session->s_cap_gen &&
1086             time_before(ttl, di->time))
1087                 return;  /* we already have a newer lease. */
1088
1089         if (di->lease_session && di->lease_session != session) {
1090                 *old_lease_session = di->lease_session;
1091                 di->lease_session = NULL;
1092         }
1093
1094         if (!di->lease_session)
1095                 di->lease_session = ceph_get_mds_session(session);
1096         di->lease_gen = session->s_cap_gen;
1097         di->lease_seq = le32_to_cpu(lease->seq);
1098         di->lease_renew_after = half_ttl;
1099         di->lease_renew_from = 0;
1100         di->time = ttl;
1101
1102         __ceph_dentry_lease_touch(di);
1103 }
1104
1105 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1106                                         struct ceph_mds_reply_lease *lease,
1107                                         struct ceph_mds_session *session,
1108                                         unsigned long from_time)
1109 {
1110         struct ceph_mds_session *old_lease_session = NULL;
1111         spin_lock(&dentry->d_lock);
1112         __update_dentry_lease(dir, dentry, lease, session, from_time,
1113                               &old_lease_session);
1114         spin_unlock(&dentry->d_lock);
1115         if (old_lease_session)
1116                 ceph_put_mds_session(old_lease_session);
1117 }
1118
1119 /*
1120  * update dentry lease without having parent inode locked
1121  */
1122 static void update_dentry_lease_careful(struct dentry *dentry,
1123                                         struct ceph_mds_reply_lease *lease,
1124                                         struct ceph_mds_session *session,
1125                                         unsigned long from_time,
1126                                         char *dname, u32 dname_len,
1127                                         struct ceph_vino *pdvino,
1128                                         struct ceph_vino *ptvino)
1129
1130 {
1131         struct inode *dir;
1132         struct ceph_mds_session *old_lease_session = NULL;
1133
1134         spin_lock(&dentry->d_lock);
1135         /* make sure dentry's name matches target */
1136         if (dentry->d_name.len != dname_len ||
1137             memcmp(dentry->d_name.name, dname, dname_len))
1138                 goto out_unlock;
1139
1140         dir = d_inode(dentry->d_parent);
1141         /* make sure parent matches dvino */
1142         if (!ceph_ino_compare(dir, pdvino))
1143                 goto out_unlock;
1144
1145         /* make sure dentry's inode matches target. NULL ptvino means that
1146          * we expect a negative dentry */
1147         if (ptvino) {
1148                 if (d_really_is_negative(dentry))
1149                         goto out_unlock;
1150                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1151                         goto out_unlock;
1152         } else {
1153                 if (d_really_is_positive(dentry))
1154                         goto out_unlock;
1155         }
1156
1157         __update_dentry_lease(dir, dentry, lease, session,
1158                               from_time, &old_lease_session);
1159 out_unlock:
1160         spin_unlock(&dentry->d_lock);
1161         if (old_lease_session)
1162                 ceph_put_mds_session(old_lease_session);
1163 }
1164
1165 /*
1166  * splice a dentry to an inode.
1167  * caller must hold directory i_mutex for this to be safe.
1168  */
1169 static int splice_dentry(struct dentry **pdn, struct inode *in)
1170 {
1171         struct dentry *dn = *pdn;
1172         struct dentry *realdn;
1173
1174         BUG_ON(d_inode(dn));
1175
1176         if (S_ISDIR(in->i_mode)) {
1177                 /* If inode is directory, d_splice_alias() below will remove
1178                  * 'realdn' from its origin parent. We need to ensure that
1179                  * origin parent's readdir cache will not reference 'realdn'
1180                  */
1181                 realdn = d_find_any_alias(in);
1182                 if (realdn) {
1183                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1184                         spin_lock(&realdn->d_lock);
1185
1186                         realdn->d_op->d_prune(realdn);
1187
1188                         di->time = jiffies;
1189                         di->lease_shared_gen = 0;
1190                         di->offset = 0;
1191
1192                         spin_unlock(&realdn->d_lock);
1193                         dput(realdn);
1194                 }
1195         }
1196
1197         /* dn must be unhashed */
1198         if (!d_unhashed(dn))
1199                 d_drop(dn);
1200         realdn = d_splice_alias(in, dn);
1201         if (IS_ERR(realdn)) {
1202                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1203                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1204                 return PTR_ERR(realdn);
1205         }
1206
1207         if (realdn) {
1208                 dout("dn %p (%d) spliced with %p (%d) "
1209                      "inode %p ino %llx.%llx\n",
1210                      dn, d_count(dn),
1211                      realdn, d_count(realdn),
1212                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1213                 dput(dn);
1214                 *pdn = realdn;
1215         } else {
1216                 BUG_ON(!ceph_dentry(dn));
1217                 dout("dn %p attached to %p ino %llx.%llx\n",
1218                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1219         }
1220         return 0;
1221 }
1222
1223 /*
1224  * Incorporate results into the local cache.  This is either just
1225  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1226  * after a lookup).
1227  *
1228  * A reply may contain
1229  *         a directory inode along with a dentry.
1230  *  and/or a target inode
1231  *
1232  * Called with snap_rwsem (read).
1233  */
1234 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1235 {
1236         struct ceph_mds_session *session = req->r_session;
1237         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1238         struct inode *in = NULL;
1239         struct ceph_vino tvino, dvino;
1240         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1241         int err = 0;
1242
1243         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1244              rinfo->head->is_dentry, rinfo->head->is_target);
1245
1246         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1247                 dout("fill_trace reply is empty!\n");
1248                 if (rinfo->head->result == 0 && req->r_parent)
1249                         ceph_invalidate_dir_request(req);
1250                 return 0;
1251         }
1252
1253         if (rinfo->head->is_dentry) {
1254                 struct inode *dir = req->r_parent;
1255
1256                 if (dir) {
1257                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1258                                               rinfo->dirfrag, session, -1,
1259                                               &req->r_caps_reservation);
1260                         if (err < 0)
1261                                 goto done;
1262                 } else {
1263                         WARN_ON_ONCE(1);
1264                 }
1265
1266                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1267                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1268                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1269                         struct qstr dname;
1270                         struct dentry *dn, *parent;
1271
1272                         BUG_ON(!rinfo->head->is_target);
1273                         BUG_ON(req->r_dentry);
1274
1275                         parent = d_find_any_alias(dir);
1276                         BUG_ON(!parent);
1277
1278                         dname.name = rinfo->dname;
1279                         dname.len = rinfo->dname_len;
1280                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1281                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1282                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1283 retry_lookup:
1284                         dn = d_lookup(parent, &dname);
1285                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1286                              parent, dname.len, dname.name, dn);
1287
1288                         if (!dn) {
1289                                 dn = d_alloc(parent, &dname);
1290                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1291                                      dname.len, dname.name, dn);
1292                                 if (!dn) {
1293                                         dput(parent);
1294                                         err = -ENOMEM;
1295                                         goto done;
1296                                 }
1297                                 err = 0;
1298                         } else if (d_really_is_positive(dn) &&
1299                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1300                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1301                                 dout(" dn %p points to wrong inode %p\n",
1302                                      dn, d_inode(dn));
1303                                 ceph_dir_clear_ordered(dir);
1304                                 d_delete(dn);
1305                                 dput(dn);
1306                                 goto retry_lookup;
1307                         }
1308
1309                         req->r_dentry = dn;
1310                         dput(parent);
1311                 }
1312         }
1313
1314         if (rinfo->head->is_target) {
1315                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1316                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1317
1318                 in = ceph_get_inode(sb, tvino);
1319                 if (IS_ERR(in)) {
1320                         err = PTR_ERR(in);
1321                         goto done;
1322                 }
1323
1324                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1325                                 NULL, session,
1326                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1327                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1328                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1329                                 &req->r_caps_reservation);
1330                 if (err < 0) {
1331                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1332                                 in, ceph_vinop(in));
1333                         if (in->i_state & I_NEW)
1334                                 discard_new_inode(in);
1335                         goto done;
1336                 }
1337                 req->r_target_inode = in;
1338                 if (in->i_state & I_NEW)
1339                         unlock_new_inode(in);
1340         }
1341
1342         /*
1343          * ignore null lease/binding on snapdir ENOENT, or else we
1344          * will have trouble splicing in the virtual snapdir later
1345          */
1346         if (rinfo->head->is_dentry &&
1347             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1348             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1349             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1350                                                fsc->mount_options->snapdir_name,
1351                                                req->r_dentry->d_name.len))) {
1352                 /*
1353                  * lookup link rename   : null -> possibly existing inode
1354                  * mknod symlink mkdir  : null -> new inode
1355                  * unlink               : linked -> null
1356                  */
1357                 struct inode *dir = req->r_parent;
1358                 struct dentry *dn = req->r_dentry;
1359                 bool have_dir_cap, have_lease;
1360
1361                 BUG_ON(!dn);
1362                 BUG_ON(!dir);
1363                 BUG_ON(d_inode(dn->d_parent) != dir);
1364
1365                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1366                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1367
1368                 BUG_ON(ceph_ino(dir) != dvino.ino);
1369                 BUG_ON(ceph_snap(dir) != dvino.snap);
1370
1371                 /* do we have a lease on the whole dir? */
1372                 have_dir_cap =
1373                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1374                          CEPH_CAP_FILE_SHARED);
1375
1376                 /* do we have a dn lease? */
1377                 have_lease = have_dir_cap ||
1378                         le32_to_cpu(rinfo->dlease->duration_ms);
1379                 if (!have_lease)
1380                         dout("fill_trace  no dentry lease or dir cap\n");
1381
1382                 /* rename? */
1383                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1384                         struct inode *olddir = req->r_old_dentry_dir;
1385                         BUG_ON(!olddir);
1386
1387                         dout(" src %p '%pd' dst %p '%pd'\n",
1388                              req->r_old_dentry,
1389                              req->r_old_dentry,
1390                              dn, dn);
1391                         dout("fill_trace doing d_move %p -> %p\n",
1392                              req->r_old_dentry, dn);
1393
1394                         /* d_move screws up sibling dentries' offsets */
1395                         ceph_dir_clear_ordered(dir);
1396                         ceph_dir_clear_ordered(olddir);
1397
1398                         d_move(req->r_old_dentry, dn);
1399                         dout(" src %p '%pd' dst %p '%pd'\n",
1400                              req->r_old_dentry,
1401                              req->r_old_dentry,
1402                              dn, dn);
1403
1404                         /* ensure target dentry is invalidated, despite
1405                            rehashing bug in vfs_rename_dir */
1406                         ceph_invalidate_dentry_lease(dn);
1407
1408                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1409                              ceph_dentry(req->r_old_dentry)->offset);
1410
1411                         /* swap r_dentry and r_old_dentry in case that
1412                          * splice_dentry() gets called later. This is safe
1413                          * because no other place will use them */
1414                         req->r_dentry = req->r_old_dentry;
1415                         req->r_old_dentry = dn;
1416                         dn = req->r_dentry;
1417                 }
1418
1419                 /* null dentry? */
1420                 if (!rinfo->head->is_target) {
1421                         dout("fill_trace null dentry\n");
1422                         if (d_really_is_positive(dn)) {
1423                                 dout("d_delete %p\n", dn);
1424                                 ceph_dir_clear_ordered(dir);
1425                                 d_delete(dn);
1426                         } else if (have_lease) {
1427                                 if (d_unhashed(dn))
1428                                         d_add(dn, NULL);
1429                                 update_dentry_lease(dir, dn,
1430                                                     rinfo->dlease, session,
1431                                                     req->r_request_started);
1432                         }
1433                         goto done;
1434                 }
1435
1436                 /* attach proper inode */
1437                 if (d_really_is_negative(dn)) {
1438                         ceph_dir_clear_ordered(dir);
1439                         ihold(in);
1440                         err = splice_dentry(&req->r_dentry, in);
1441                         if (err < 0)
1442                                 goto done;
1443                         dn = req->r_dentry;  /* may have spliced */
1444                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1445                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1446                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1447                              ceph_vinop(in));
1448                         d_invalidate(dn);
1449                         have_lease = false;
1450                 }
1451
1452                 if (have_lease) {
1453                         update_dentry_lease(dir, dn,
1454                                             rinfo->dlease, session,
1455                                             req->r_request_started);
1456                 }
1457                 dout(" final dn %p\n", dn);
1458         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1459                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1460                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1461                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1462                 struct inode *dir = req->r_parent;
1463
1464                 /* fill out a snapdir LOOKUPSNAP dentry */
1465                 BUG_ON(!dir);
1466                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1467                 BUG_ON(!req->r_dentry);
1468                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1469                 ceph_dir_clear_ordered(dir);
1470                 ihold(in);
1471                 err = splice_dentry(&req->r_dentry, in);
1472                 if (err < 0)
1473                         goto done;
1474         } else if (rinfo->head->is_dentry && req->r_dentry) {
1475                 /* parent inode is not locked, be carefull */
1476                 struct ceph_vino *ptvino = NULL;
1477                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1478                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1479                 if (rinfo->head->is_target) {
1480                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1481                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1482                         ptvino = &tvino;
1483                 }
1484                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1485                                             session, req->r_request_started,
1486                                             rinfo->dname, rinfo->dname_len,
1487                                             &dvino, ptvino);
1488         }
1489 done:
1490         dout("fill_trace done err=%d\n", err);
1491         return err;
1492 }
1493
1494 /*
1495  * Prepopulate our cache with readdir results, leases, etc.
1496  */
1497 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1498                                            struct ceph_mds_session *session)
1499 {
1500         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1501         int i, err = 0;
1502
1503         for (i = 0; i < rinfo->dir_nr; i++) {
1504                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1505                 struct ceph_vino vino;
1506                 struct inode *in;
1507                 int rc;
1508
1509                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1510                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1511
1512                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1513                 if (IS_ERR(in)) {
1514                         err = PTR_ERR(in);
1515                         dout("new_inode badness got %d\n", err);
1516                         continue;
1517                 }
1518                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1519                                      -1, &req->r_caps_reservation);
1520                 if (rc < 0) {
1521                         pr_err("ceph_fill_inode badness on %p got %d\n",
1522                                in, rc);
1523                         err = rc;
1524                         if (in->i_state & I_NEW) {
1525                                 ihold(in);
1526                                 discard_new_inode(in);
1527                         }
1528                 } else if (in->i_state & I_NEW) {
1529                         unlock_new_inode(in);
1530                 }
1531
1532                 /* avoid calling iput_final() in mds dispatch threads */
1533                 ceph_async_iput(in);
1534         }
1535
1536         return err;
1537 }
1538
1539 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1540 {
1541         if (ctl->page) {
1542                 kunmap(ctl->page);
1543                 put_page(ctl->page);
1544                 ctl->page = NULL;
1545         }
1546 }
1547
1548 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1549                               struct ceph_readdir_cache_control *ctl,
1550                               struct ceph_mds_request *req)
1551 {
1552         struct ceph_inode_info *ci = ceph_inode(dir);
1553         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1554         unsigned idx = ctl->index % nsize;
1555         pgoff_t pgoff = ctl->index / nsize;
1556
1557         if (!ctl->page || pgoff != page_index(ctl->page)) {
1558                 ceph_readdir_cache_release(ctl);
1559                 if (idx == 0)
1560                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1561                 else
1562                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1563                 if (!ctl->page) {
1564                         ctl->index = -1;
1565                         return idx == 0 ? -ENOMEM : 0;
1566                 }
1567                 /* reading/filling the cache are serialized by
1568                  * i_mutex, no need to use page lock */
1569                 unlock_page(ctl->page);
1570                 ctl->dentries = kmap(ctl->page);
1571                 if (idx == 0)
1572                         memset(ctl->dentries, 0, PAGE_SIZE);
1573         }
1574
1575         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1576             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1577                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1578                 ctl->dentries[idx] = dn;
1579                 ctl->index++;
1580         } else {
1581                 dout("disable readdir cache\n");
1582                 ctl->index = -1;
1583         }
1584         return 0;
1585 }
1586
1587 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1588                              struct ceph_mds_session *session)
1589 {
1590         struct dentry *parent = req->r_dentry;
1591         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1592         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1593         struct qstr dname;
1594         struct dentry *dn;
1595         struct inode *in;
1596         int err = 0, skipped = 0, ret, i;
1597         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1598         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1599         u32 last_hash = 0;
1600         u32 fpos_offset;
1601         struct ceph_readdir_cache_control cache_ctl = {};
1602
1603         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1604                 return readdir_prepopulate_inodes_only(req, session);
1605
1606         if (rinfo->hash_order) {
1607                 if (req->r_path2) {
1608                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1609                                                   req->r_path2,
1610                                                   strlen(req->r_path2));
1611                         last_hash = ceph_frag_value(last_hash);
1612                 } else if (rinfo->offset_hash) {
1613                         /* mds understands offset_hash */
1614                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1615                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1616                 }
1617         }
1618
1619         if (rinfo->dir_dir &&
1620             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1621                 dout("readdir_prepopulate got new frag %x -> %x\n",
1622                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1623                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1624                 if (!rinfo->hash_order)
1625                         req->r_readdir_offset = 2;
1626         }
1627
1628         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1629                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1630                      rinfo->dir_nr, parent);
1631         } else {
1632                 dout("readdir_prepopulate %d items under dn %p\n",
1633                      rinfo->dir_nr, parent);
1634                 if (rinfo->dir_dir)
1635                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1636
1637                 if (ceph_frag_is_leftmost(frag) &&
1638                     req->r_readdir_offset == 2 &&
1639                     !(rinfo->hash_order && last_hash)) {
1640                         /* note dir version at start of readdir so we can
1641                          * tell if any dentries get dropped */
1642                         req->r_dir_release_cnt =
1643                                 atomic64_read(&ci->i_release_count);
1644                         req->r_dir_ordered_cnt =
1645                                 atomic64_read(&ci->i_ordered_count);
1646                         req->r_readdir_cache_idx = 0;
1647                 }
1648         }
1649
1650         cache_ctl.index = req->r_readdir_cache_idx;
1651         fpos_offset = req->r_readdir_offset;
1652
1653         /* FIXME: release caps/leases if error occurs */
1654         for (i = 0; i < rinfo->dir_nr; i++) {
1655                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1656                 struct ceph_vino tvino;
1657
1658                 dname.name = rde->name;
1659                 dname.len = rde->name_len;
1660                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1661
1662                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1663                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1664
1665                 if (rinfo->hash_order) {
1666                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1667                                                  rde->name, rde->name_len);
1668                         hash = ceph_frag_value(hash);
1669                         if (hash != last_hash)
1670                                 fpos_offset = 2;
1671                         last_hash = hash;
1672                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1673                 } else {
1674                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1675                 }
1676
1677 retry_lookup:
1678                 dn = d_lookup(parent, &dname);
1679                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1680                      parent, dname.len, dname.name, dn);
1681
1682                 if (!dn) {
1683                         dn = d_alloc(parent, &dname);
1684                         dout("d_alloc %p '%.*s' = %p\n", parent,
1685                              dname.len, dname.name, dn);
1686                         if (!dn) {
1687                                 dout("d_alloc badness\n");
1688                                 err = -ENOMEM;
1689                                 goto out;
1690                         }
1691                 } else if (d_really_is_positive(dn) &&
1692                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1693                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1694                         struct ceph_dentry_info *di = ceph_dentry(dn);
1695                         dout(" dn %p points to wrong inode %p\n",
1696                              dn, d_inode(dn));
1697
1698                         spin_lock(&dn->d_lock);
1699                         if (di->offset > 0 &&
1700                             di->lease_shared_gen ==
1701                             atomic_read(&ci->i_shared_gen)) {
1702                                 __ceph_dir_clear_ordered(ci);
1703                                 di->offset = 0;
1704                         }
1705                         spin_unlock(&dn->d_lock);
1706
1707                         d_delete(dn);
1708                         dput(dn);
1709                         goto retry_lookup;
1710                 }
1711
1712                 /* inode */
1713                 if (d_really_is_positive(dn)) {
1714                         in = d_inode(dn);
1715                 } else {
1716                         in = ceph_get_inode(parent->d_sb, tvino);
1717                         if (IS_ERR(in)) {
1718                                 dout("new_inode badness\n");
1719                                 d_drop(dn);
1720                                 dput(dn);
1721                                 err = PTR_ERR(in);
1722                                 goto out;
1723                         }
1724                 }
1725
1726                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1727                                       -1, &req->r_caps_reservation);
1728                 if (ret < 0) {
1729                         pr_err("ceph_fill_inode badness on %p\n", in);
1730                         if (d_really_is_negative(dn)) {
1731                                 /* avoid calling iput_final() in mds
1732                                  * dispatch threads */
1733                                 if (in->i_state & I_NEW) {
1734                                         ihold(in);
1735                                         discard_new_inode(in);
1736                                 }
1737                                 ceph_async_iput(in);
1738                         }
1739                         d_drop(dn);
1740                         err = ret;
1741                         goto next_item;
1742                 }
1743                 if (in->i_state & I_NEW)
1744                         unlock_new_inode(in);
1745
1746                 if (d_really_is_negative(dn)) {
1747                         if (ceph_security_xattr_deadlock(in)) {
1748                                 dout(" skip splicing dn %p to inode %p"
1749                                      " (security xattr deadlock)\n", dn, in);
1750                                 ceph_async_iput(in);
1751                                 skipped++;
1752                                 goto next_item;
1753                         }
1754
1755                         err = splice_dentry(&dn, in);
1756                         if (err < 0)
1757                                 goto next_item;
1758                 }
1759
1760                 ceph_dentry(dn)->offset = rde->offset;
1761
1762                 update_dentry_lease(d_inode(parent), dn,
1763                                     rde->lease, req->r_session,
1764                                     req->r_request_started);
1765
1766                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1767                         ret = fill_readdir_cache(d_inode(parent), dn,
1768                                                  &cache_ctl, req);
1769                         if (ret < 0)
1770                                 err = ret;
1771                 }
1772 next_item:
1773                 dput(dn);
1774         }
1775 out:
1776         if (err == 0 && skipped == 0) {
1777                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1778                 req->r_readdir_cache_idx = cache_ctl.index;
1779         }
1780         ceph_readdir_cache_release(&cache_ctl);
1781         dout("readdir_prepopulate done\n");
1782         return err;
1783 }
1784
1785 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1786 {
1787         struct ceph_inode_info *ci = ceph_inode(inode);
1788         bool ret;
1789
1790         spin_lock(&ci->i_ceph_lock);
1791         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1792         i_size_write(inode, size);
1793         inode->i_blocks = calc_inode_blocks(size);
1794
1795         ret = __ceph_should_report_size(ci);
1796
1797         spin_unlock(&ci->i_ceph_lock);
1798         return ret;
1799 }
1800
1801 /*
1802  * Put reference to inode, but avoid calling iput_final() in current thread.
1803  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1804  * some contexts.
1805  */
1806 void ceph_async_iput(struct inode *inode)
1807 {
1808         if (!inode)
1809                 return;
1810         for (;;) {
1811                 if (atomic_add_unless(&inode->i_count, -1, 1))
1812                         break;
1813                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1814                                &ceph_inode(inode)->i_work))
1815                         break;
1816                 /* queue work failed, i_count must be at least 2 */
1817         }
1818 }
1819
1820 /*
1821  * Write back inode data in a worker thread.  (This can't be done
1822  * in the message handler context.)
1823  */
1824 void ceph_queue_writeback(struct inode *inode)
1825 {
1826         struct ceph_inode_info *ci = ceph_inode(inode);
1827         set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1828
1829         ihold(inode);
1830         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1831                        &ci->i_work)) {
1832                 dout("ceph_queue_writeback %p\n", inode);
1833         } else {
1834                 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1835                      inode, ci->i_work_mask);
1836                 iput(inode);
1837         }
1838 }
1839
1840 /*
1841  * queue an async invalidation
1842  */
1843 void ceph_queue_invalidate(struct inode *inode)
1844 {
1845         struct ceph_inode_info *ci = ceph_inode(inode);
1846         set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1847
1848         ihold(inode);
1849         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1850                        &ceph_inode(inode)->i_work)) {
1851                 dout("ceph_queue_invalidate %p\n", inode);
1852         } else {
1853                 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1854                      inode, ci->i_work_mask);
1855                 iput(inode);
1856         }
1857 }
1858
1859 /*
1860  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1861  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1862  */
1863 void ceph_queue_vmtruncate(struct inode *inode)
1864 {
1865         struct ceph_inode_info *ci = ceph_inode(inode);
1866         set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1867
1868         ihold(inode);
1869         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1870                        &ci->i_work)) {
1871                 dout("ceph_queue_vmtruncate %p\n", inode);
1872         } else {
1873                 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1874                      inode, ci->i_work_mask);
1875                 iput(inode);
1876         }
1877 }
1878
1879 static void ceph_do_invalidate_pages(struct inode *inode)
1880 {
1881         struct ceph_inode_info *ci = ceph_inode(inode);
1882         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1883         u32 orig_gen;
1884         int check = 0;
1885
1886         mutex_lock(&ci->i_truncate_mutex);
1887
1888         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1889                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1890                                     inode, ceph_ino(inode));
1891                 mapping_set_error(inode->i_mapping, -EIO);
1892                 truncate_pagecache(inode, 0);
1893                 mutex_unlock(&ci->i_truncate_mutex);
1894                 goto out;
1895         }
1896
1897         spin_lock(&ci->i_ceph_lock);
1898         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1899              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1900         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1901                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1902                         check = 1;
1903                 spin_unlock(&ci->i_ceph_lock);
1904                 mutex_unlock(&ci->i_truncate_mutex);
1905                 goto out;
1906         }
1907         orig_gen = ci->i_rdcache_gen;
1908         spin_unlock(&ci->i_ceph_lock);
1909
1910         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1911                 pr_err("invalidate_pages %p fails\n", inode);
1912         }
1913
1914         spin_lock(&ci->i_ceph_lock);
1915         if (orig_gen == ci->i_rdcache_gen &&
1916             orig_gen == ci->i_rdcache_revoking) {
1917                 dout("invalidate_pages %p gen %d successful\n", inode,
1918                      ci->i_rdcache_gen);
1919                 ci->i_rdcache_revoking--;
1920                 check = 1;
1921         } else {
1922                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1923                      inode, orig_gen, ci->i_rdcache_gen,
1924                      ci->i_rdcache_revoking);
1925                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1926                         check = 1;
1927         }
1928         spin_unlock(&ci->i_ceph_lock);
1929         mutex_unlock(&ci->i_truncate_mutex);
1930 out:
1931         if (check)
1932                 ceph_check_caps(ci, 0, NULL);
1933 }
1934
1935 /*
1936  * Make sure any pending truncation is applied before doing anything
1937  * that may depend on it.
1938  */
1939 void __ceph_do_pending_vmtruncate(struct inode *inode)
1940 {
1941         struct ceph_inode_info *ci = ceph_inode(inode);
1942         u64 to;
1943         int wrbuffer_refs, finish = 0;
1944
1945         mutex_lock(&ci->i_truncate_mutex);
1946 retry:
1947         spin_lock(&ci->i_ceph_lock);
1948         if (ci->i_truncate_pending == 0) {
1949                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1950                 spin_unlock(&ci->i_ceph_lock);
1951                 mutex_unlock(&ci->i_truncate_mutex);
1952                 return;
1953         }
1954
1955         /*
1956          * make sure any dirty snapped pages are flushed before we
1957          * possibly truncate them.. so write AND block!
1958          */
1959         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1960                 spin_unlock(&ci->i_ceph_lock);
1961                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1962                      inode);
1963                 filemap_write_and_wait_range(&inode->i_data, 0,
1964                                              inode->i_sb->s_maxbytes);
1965                 goto retry;
1966         }
1967
1968         /* there should be no reader or writer */
1969         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1970
1971         to = ci->i_truncate_size;
1972         wrbuffer_refs = ci->i_wrbuffer_ref;
1973         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1974              ci->i_truncate_pending, to);
1975         spin_unlock(&ci->i_ceph_lock);
1976
1977         truncate_pagecache(inode, to);
1978
1979         spin_lock(&ci->i_ceph_lock);
1980         if (to == ci->i_truncate_size) {
1981                 ci->i_truncate_pending = 0;
1982                 finish = 1;
1983         }
1984         spin_unlock(&ci->i_ceph_lock);
1985         if (!finish)
1986                 goto retry;
1987
1988         mutex_unlock(&ci->i_truncate_mutex);
1989
1990         if (wrbuffer_refs == 0)
1991                 ceph_check_caps(ci, 0, NULL);
1992
1993         wake_up_all(&ci->i_cap_wq);
1994 }
1995
1996 static void ceph_inode_work(struct work_struct *work)
1997 {
1998         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1999                                                  i_work);
2000         struct inode *inode = &ci->vfs_inode;
2001
2002         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2003                 dout("writeback %p\n", inode);
2004                 filemap_fdatawrite(&inode->i_data);
2005         }
2006         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2007                 ceph_do_invalidate_pages(inode);
2008
2009         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2010                 __ceph_do_pending_vmtruncate(inode);
2011
2012         iput(inode);
2013 }
2014
2015 /*
2016  * symlinks
2017  */
2018 static const struct inode_operations ceph_symlink_iops = {
2019         .get_link = simple_get_link,
2020         .setattr = ceph_setattr,
2021         .getattr = ceph_getattr,
2022         .listxattr = ceph_listxattr,
2023 };
2024
2025 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2026 {
2027         struct ceph_inode_info *ci = ceph_inode(inode);
2028         unsigned int ia_valid = attr->ia_valid;
2029         struct ceph_mds_request *req;
2030         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2031         struct ceph_cap_flush *prealloc_cf;
2032         int issued;
2033         int release = 0, dirtied = 0;
2034         int mask = 0;
2035         int err = 0;
2036         int inode_dirty_flags = 0;
2037         bool lock_snap_rwsem = false;
2038
2039         prealloc_cf = ceph_alloc_cap_flush();
2040         if (!prealloc_cf)
2041                 return -ENOMEM;
2042
2043         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2044                                        USE_AUTH_MDS);
2045         if (IS_ERR(req)) {
2046                 ceph_free_cap_flush(prealloc_cf);
2047                 return PTR_ERR(req);
2048         }
2049
2050         spin_lock(&ci->i_ceph_lock);
2051         issued = __ceph_caps_issued(ci, NULL);
2052
2053         if (!ci->i_head_snapc &&
2054             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2055                 lock_snap_rwsem = true;
2056                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2057                         spin_unlock(&ci->i_ceph_lock);
2058                         down_read(&mdsc->snap_rwsem);
2059                         spin_lock(&ci->i_ceph_lock);
2060                         issued = __ceph_caps_issued(ci, NULL);
2061                 }
2062         }
2063
2064         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2065
2066         if (ia_valid & ATTR_UID) {
2067                 dout("setattr %p uid %d -> %d\n", inode,
2068                      from_kuid(&init_user_ns, inode->i_uid),
2069                      from_kuid(&init_user_ns, attr->ia_uid));
2070                 if (issued & CEPH_CAP_AUTH_EXCL) {
2071                         inode->i_uid = attr->ia_uid;
2072                         dirtied |= CEPH_CAP_AUTH_EXCL;
2073                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2074                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2075                         req->r_args.setattr.uid = cpu_to_le32(
2076                                 from_kuid(&init_user_ns, attr->ia_uid));
2077                         mask |= CEPH_SETATTR_UID;
2078                         release |= CEPH_CAP_AUTH_SHARED;
2079                 }
2080         }
2081         if (ia_valid & ATTR_GID) {
2082                 dout("setattr %p gid %d -> %d\n", inode,
2083                      from_kgid(&init_user_ns, inode->i_gid),
2084                      from_kgid(&init_user_ns, attr->ia_gid));
2085                 if (issued & CEPH_CAP_AUTH_EXCL) {
2086                         inode->i_gid = attr->ia_gid;
2087                         dirtied |= CEPH_CAP_AUTH_EXCL;
2088                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2089                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2090                         req->r_args.setattr.gid = cpu_to_le32(
2091                                 from_kgid(&init_user_ns, attr->ia_gid));
2092                         mask |= CEPH_SETATTR_GID;
2093                         release |= CEPH_CAP_AUTH_SHARED;
2094                 }
2095         }
2096         if (ia_valid & ATTR_MODE) {
2097                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2098                      attr->ia_mode);
2099                 if (issued & CEPH_CAP_AUTH_EXCL) {
2100                         inode->i_mode = attr->ia_mode;
2101                         dirtied |= CEPH_CAP_AUTH_EXCL;
2102                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2103                            attr->ia_mode != inode->i_mode) {
2104                         inode->i_mode = attr->ia_mode;
2105                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2106                         mask |= CEPH_SETATTR_MODE;
2107                         release |= CEPH_CAP_AUTH_SHARED;
2108                 }
2109         }
2110
2111         if (ia_valid & ATTR_ATIME) {
2112                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2113                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2114                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2115                 if (issued & CEPH_CAP_FILE_EXCL) {
2116                         ci->i_time_warp_seq++;
2117                         inode->i_atime = attr->ia_atime;
2118                         dirtied |= CEPH_CAP_FILE_EXCL;
2119                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2120                            timespec64_compare(&inode->i_atime,
2121                                             &attr->ia_atime) < 0) {
2122                         inode->i_atime = attr->ia_atime;
2123                         dirtied |= CEPH_CAP_FILE_WR;
2124                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2125                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2126                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2127                                                &attr->ia_atime);
2128                         mask |= CEPH_SETATTR_ATIME;
2129                         release |= CEPH_CAP_FILE_SHARED |
2130                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2131                 }
2132         }
2133         if (ia_valid & ATTR_SIZE) {
2134                 dout("setattr %p size %lld -> %lld\n", inode,
2135                      inode->i_size, attr->ia_size);
2136                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2137                     attr->ia_size > inode->i_size) {
2138                         i_size_write(inode, attr->ia_size);
2139                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2140                         ci->i_reported_size = attr->ia_size;
2141                         dirtied |= CEPH_CAP_FILE_EXCL;
2142                         ia_valid |= ATTR_MTIME;
2143                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2144                            attr->ia_size != inode->i_size) {
2145                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2146                         req->r_args.setattr.old_size =
2147                                 cpu_to_le64(inode->i_size);
2148                         mask |= CEPH_SETATTR_SIZE;
2149                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2150                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2151                 }
2152         }
2153         if (ia_valid & ATTR_MTIME) {
2154                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2155                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2156                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2157                 if (issued & CEPH_CAP_FILE_EXCL) {
2158                         ci->i_time_warp_seq++;
2159                         inode->i_mtime = attr->ia_mtime;
2160                         dirtied |= CEPH_CAP_FILE_EXCL;
2161                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2162                            timespec64_compare(&inode->i_mtime,
2163                                             &attr->ia_mtime) < 0) {
2164                         inode->i_mtime = attr->ia_mtime;
2165                         dirtied |= CEPH_CAP_FILE_WR;
2166                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2167                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2168                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2169                                                &attr->ia_mtime);
2170                         mask |= CEPH_SETATTR_MTIME;
2171                         release |= CEPH_CAP_FILE_SHARED |
2172                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2173                 }
2174         }
2175
2176         /* these do nothing */
2177         if (ia_valid & ATTR_CTIME) {
2178                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2179                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2180                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2181                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2182                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2183                      only ? "ctime only" : "ignored");
2184                 if (only) {
2185                         /*
2186                          * if kernel wants to dirty ctime but nothing else,
2187                          * we need to choose a cap to dirty under, or do
2188                          * a almost-no-op setattr
2189                          */
2190                         if (issued & CEPH_CAP_AUTH_EXCL)
2191                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2192                         else if (issued & CEPH_CAP_FILE_EXCL)
2193                                 dirtied |= CEPH_CAP_FILE_EXCL;
2194                         else if (issued & CEPH_CAP_XATTR_EXCL)
2195                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2196                         else
2197                                 mask |= CEPH_SETATTR_CTIME;
2198                 }
2199         }
2200         if (ia_valid & ATTR_FILE)
2201                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2202
2203         if (dirtied) {
2204                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2205                                                            &prealloc_cf);
2206                 inode->i_ctime = attr->ia_ctime;
2207         }
2208
2209         release &= issued;
2210         spin_unlock(&ci->i_ceph_lock);
2211         if (lock_snap_rwsem)
2212                 up_read(&mdsc->snap_rwsem);
2213
2214         if (inode_dirty_flags)
2215                 __mark_inode_dirty(inode, inode_dirty_flags);
2216
2217
2218         if (mask) {
2219                 req->r_inode = inode;
2220                 ihold(inode);
2221                 req->r_inode_drop = release;
2222                 req->r_args.setattr.mask = cpu_to_le32(mask);
2223                 req->r_num_caps = 1;
2224                 req->r_stamp = attr->ia_ctime;
2225                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2226         }
2227         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2228              ceph_cap_string(dirtied), mask);
2229
2230         ceph_mdsc_put_request(req);
2231         ceph_free_cap_flush(prealloc_cf);
2232
2233         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2234                 __ceph_do_pending_vmtruncate(inode);
2235
2236         return err;
2237 }
2238
2239 /*
2240  * setattr
2241  */
2242 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2243 {
2244         struct inode *inode = d_inode(dentry);
2245         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2246         int err;
2247
2248         if (ceph_snap(inode) != CEPH_NOSNAP)
2249                 return -EROFS;
2250
2251         err = setattr_prepare(dentry, attr);
2252         if (err != 0)
2253                 return err;
2254
2255         if ((attr->ia_valid & ATTR_SIZE) &&
2256             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2257                 return -EFBIG;
2258
2259         if ((attr->ia_valid & ATTR_SIZE) &&
2260             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2261                 return -EDQUOT;
2262
2263         err = __ceph_setattr(inode, attr);
2264
2265         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2266                 err = posix_acl_chmod(inode, attr->ia_mode);
2267
2268         return err;
2269 }
2270
2271 /*
2272  * Verify that we have a lease on the given mask.  If not,
2273  * do a getattr against an mds.
2274  */
2275 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2276                       int mask, bool force)
2277 {
2278         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2279         struct ceph_mds_client *mdsc = fsc->mdsc;
2280         struct ceph_mds_request *req;
2281         int mode;
2282         int err;
2283
2284         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2285                 dout("do_getattr inode %p SNAPDIR\n", inode);
2286                 return 0;
2287         }
2288
2289         dout("do_getattr inode %p mask %s mode 0%o\n",
2290              inode, ceph_cap_string(mask), inode->i_mode);
2291         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2292                         return 0;
2293
2294         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2295         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2296         if (IS_ERR(req))
2297                 return PTR_ERR(req);
2298         req->r_inode = inode;
2299         ihold(inode);
2300         req->r_num_caps = 1;
2301         req->r_args.getattr.mask = cpu_to_le32(mask);
2302         req->r_locked_page = locked_page;
2303         err = ceph_mdsc_do_request(mdsc, NULL, req);
2304         if (locked_page && err == 0) {
2305                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2306                 if (inline_version == 0) {
2307                         /* the reply is supposed to contain inline data */
2308                         err = -EINVAL;
2309                 } else if (inline_version == CEPH_INLINE_NONE) {
2310                         err = -ENODATA;
2311                 } else {
2312                         err = req->r_reply_info.targeti.inline_len;
2313                 }
2314         }
2315         ceph_mdsc_put_request(req);
2316         dout("do_getattr result=%d\n", err);
2317         return err;
2318 }
2319
2320
2321 /*
2322  * Check inode permissions.  We verify we have a valid value for
2323  * the AUTH cap, then call the generic handler.
2324  */
2325 int ceph_permission(struct inode *inode, int mask)
2326 {
2327         int err;
2328
2329         if (mask & MAY_NOT_BLOCK)
2330                 return -ECHILD;
2331
2332         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2333
2334         if (!err)
2335                 err = generic_permission(inode, mask);
2336         return err;
2337 }
2338
2339 /* Craft a mask of needed caps given a set of requested statx attrs. */
2340 static int statx_to_caps(u32 want)
2341 {
2342         int mask = 0;
2343
2344         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2345                 mask |= CEPH_CAP_AUTH_SHARED;
2346
2347         if (want & (STATX_NLINK|STATX_CTIME))
2348                 mask |= CEPH_CAP_LINK_SHARED;
2349
2350         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2351                     STATX_BLOCKS))
2352                 mask |= CEPH_CAP_FILE_SHARED;
2353
2354         if (want & (STATX_CTIME))
2355                 mask |= CEPH_CAP_XATTR_SHARED;
2356
2357         return mask;
2358 }
2359
2360 /*
2361  * Get all the attributes. If we have sufficient caps for the requested attrs,
2362  * then we can avoid talking to the MDS at all.
2363  */
2364 int ceph_getattr(const struct path *path, struct kstat *stat,
2365                  u32 request_mask, unsigned int flags)
2366 {
2367         struct inode *inode = d_inode(path->dentry);
2368         struct ceph_inode_info *ci = ceph_inode(inode);
2369         u32 valid_mask = STATX_BASIC_STATS;
2370         int err = 0;
2371
2372         /* Skip the getattr altogether if we're asked not to sync */
2373         if (!(flags & AT_STATX_DONT_SYNC)) {
2374                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2375                                       flags & AT_STATX_FORCE_SYNC);
2376                 if (err)
2377                         return err;
2378         }
2379
2380         generic_fillattr(inode, stat);
2381         stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2382
2383         /*
2384          * btime on newly-allocated inodes is 0, so if this is still set to
2385          * that, then assume that it's not valid.
2386          */
2387         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2388                 stat->btime = ci->i_btime;
2389                 valid_mask |= STATX_BTIME;
2390         }
2391
2392         if (ceph_snap(inode) == CEPH_NOSNAP)
2393                 stat->dev = inode->i_sb->s_dev;
2394         else
2395                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2396
2397         if (S_ISDIR(inode->i_mode)) {
2398                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2399                                         RBYTES))
2400                         stat->size = ci->i_rbytes;
2401                 else
2402                         stat->size = ci->i_files + ci->i_subdirs;
2403                 stat->blocks = 0;
2404                 stat->blksize = 65536;
2405                 /*
2406                  * Some applications rely on the number of st_nlink
2407                  * value on directories to be either 0 (if unlinked)
2408                  * or 2 + number of subdirectories.
2409                  */
2410                 if (stat->nlink == 1)
2411                         /* '.' + '..' + subdirs */
2412                         stat->nlink = 1 + 1 + ci->i_subdirs;
2413         }
2414
2415         stat->result_mask = request_mask & valid_mask;
2416         return err;
2417 }