fs/ceph/inode.c

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