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