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         kmem_cache_free(ceph_inode_cachep, ci);
 478 }
 479
 480 void ceph_destroy_inode(struct inode *inode)
 481 {
 482         struct ceph_inode_info *ci = ceph_inode(inode);
 483         struct ceph_inode_frag *frag;
 484         struct rb_node *n;
 485
 486         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 487
 488         ceph_fscache_unregister_inode_cookie(ci);
 489
 490         ceph_queue_caps_release(inode);
 491
 492         /*
 493          * we may still have a snap_realm reference if there are stray
 494          * caps in i_snap_caps.
 495          */
 496         if (ci->i_snap_realm) {
 497                 struct ceph_mds_client *mdsc =
 498                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 499                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 500
 501                 dout(" dropping residual ref to snap realm %p\n", realm);
 502                 spin_lock(&realm->inodes_with_caps_lock);
 503                 list_del_init(&ci->i_snap_realm_item);
 504                 spin_unlock(&realm->inodes_with_caps_lock);
 505                 ceph_put_snap_realm(mdsc, realm);
 506         }
 507
 508         kfree(ci->i_symlink);
 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                 i_size_write(inode, 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         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 729
 730         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 731             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 732                 inode->i_mode = le32_to_cpu(info->mode);
 733                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 734                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 735                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 736                      from_kuid(&init_user_ns, inode->i_uid),
 737                      from_kgid(&init_user_ns, inode->i_gid));
 738         }
 739
 740         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 741             (issued & CEPH_CAP_LINK_EXCL) == 0)
 742                 set_nlink(inode, le32_to_cpu(info->nlink));
 743
 744         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 745                 /* be careful with mtime, atime, size */
 746                 ceph_decode_timespec(&atime, &info->atime);
 747                 ceph_decode_timespec(&mtime, &info->mtime);
 748                 ceph_decode_timespec(&ctime, &info->ctime);
 749                 ceph_fill_file_time(inode, issued,
 750                                 le32_to_cpu(info->time_warp_seq),
 751                                 &ctime, &mtime, &atime);
 752         }
 753
 754         if (new_version ||
 755             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 756                 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
 757                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 758                 ci->i_layout = info->layout;
 759
 760                 queue_trunc = ceph_fill_file_size(inode, issued,
 761                                         le32_to_cpu(info->truncate_seq),
 762                                         le64_to_cpu(info->truncate_size),
 763                                         le64_to_cpu(info->size));
 764                 /* only update max_size on auth cap */
 765                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 766                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 767                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 768                                         le64_to_cpu(info->max_size));
 769                         ci->i_max_size = le64_to_cpu(info->max_size);
 770                 }
 771         }
 772
 773         /* xattrs */
 774         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 775         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 776             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 777                 if (ci->i_xattrs.blob)
 778                         ceph_buffer_put(ci->i_xattrs.blob);
 779                 ci->i_xattrs.blob = xattr_blob;
 780                 if (xattr_blob)
 781                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 782                                iinfo->xattr_data, iinfo->xattr_len);
 783                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 784                 ceph_forget_all_cached_acls(inode);
 785                 xattr_blob = NULL;
 786         }
 787
 788         inode->i_mapping->a_ops = &ceph_aops;
 789
 790         switch (inode->i_mode & S_IFMT) {
 791         case S_IFIFO:
 792         case S_IFBLK:
 793         case S_IFCHR:
 794         case S_IFSOCK:
 795                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 796                 inode->i_op = &ceph_file_iops;
 797                 break;
 798         case S_IFREG:
 799                 inode->i_op = &ceph_file_iops;
 800                 inode->i_fop = &ceph_file_fops;
 801                 break;
 802         case S_IFLNK:
 803                 inode->i_op = &ceph_symlink_iops;
 804                 if (!ci->i_symlink) {
 805                         u32 symlen = iinfo->symlink_len;
 806                         char *sym;
 807
 808                         spin_unlock(&ci->i_ceph_lock);
 809
 810                         err = -EINVAL;
 811                         if (WARN_ON(symlen != i_size_read(inode)))
 812                                 goto out;
 813
 814                         err = -ENOMEM;
 815                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 816                         if (!sym)
 817                                 goto out;
 818
 819                         spin_lock(&ci->i_ceph_lock);
 820                         if (!ci->i_symlink)
 821                                 ci->i_symlink = sym;
 822                         else
 823                                 kfree(sym); /* lost a race */
 824                 }
 825                 inode->i_link = ci->i_symlink;
 826                 break;
 827         case S_IFDIR:
 828                 inode->i_op = &ceph_dir_iops;
 829                 inode->i_fop = &ceph_dir_fops;
 830
 831                 ci->i_dir_layout = iinfo->dir_layout;
 832
 833                 ci->i_files = le64_to_cpu(info->files);
 834                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 835                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 836                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 837                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 838                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 839                 break;
 840         default:
 841                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 842                        ceph_vinop(inode), inode->i_mode);
 843         }
 844
 845         /* were we issued a capability? */
 846         if (info->cap.caps) {
 847                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 848                         unsigned caps = le32_to_cpu(info->cap.caps);
 849                         ceph_add_cap(inode, session,
 850                                      le64_to_cpu(info->cap.cap_id),
 851                                      cap_fmode, caps,
 852                                      le32_to_cpu(info->cap.wanted),
 853                                      le32_to_cpu(info->cap.seq),
 854                                      le32_to_cpu(info->cap.mseq),
 855                                      le64_to_cpu(info->cap.realm),
 856                                      info->cap.flags, &new_cap);
 857
 858                         /* set dir completion flag? */
 859                         if (S_ISDIR(inode->i_mode) &&
 860                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 861                             (caps & CEPH_CAP_FILE_SHARED) &&
 862                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 863                             !__ceph_dir_is_complete(ci)) {
 864                                 dout(" marking %p complete (empty)\n", inode);
 865                                 i_size_write(inode, 0);
 866                                 __ceph_dir_set_complete(ci,
 867                                         atomic64_read(&ci->i_release_count),
 868                                         atomic64_read(&ci->i_ordered_count));
 869                         }
 870
 871                         wake = true;
 872                 } else {
 873                         dout(" %p got snap_caps %s\n", inode,
 874                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 875                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 876                         if (cap_fmode >= 0)
 877                                 __ceph_get_fmode(ci, cap_fmode);
 878                 }
 879         } else if (cap_fmode >= 0) {
 880                 pr_warn("mds issued no caps on %llx.%llx\n",
 881                            ceph_vinop(inode));
 882                 __ceph_get_fmode(ci, cap_fmode);
 883         }
 884
 885         if (iinfo->inline_version > 0 &&
 886             iinfo->inline_version >= ci->i_inline_version) {
 887                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 888                 ci->i_inline_version = iinfo->inline_version;
 889                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 890                     (locked_page ||
 891                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 892                         fill_inline = true;
 893         }
 894
 895         spin_unlock(&ci->i_ceph_lock);
 896
 897         if (fill_inline)
 898                 ceph_fill_inline_data(inode, locked_page,
 899                                       iinfo->inline_data, iinfo->inline_len);
 900
 901         if (wake)
 902                 wake_up_all(&ci->i_cap_wq);
 903
 904         /* queue truncate if we saw i_size decrease */
 905         if (queue_trunc)
 906                 ceph_queue_vmtruncate(inode);
 907
 908         /* populate frag tree */
 909         if (S_ISDIR(inode->i_mode))
 910                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 911
 912         /* update delegation info? */
 913         if (dirinfo)
 914                 ceph_fill_dirfrag(inode, dirinfo);
 915
 916         err = 0;
 917 out:
 918         if (new_cap)
 919                 ceph_put_cap(mdsc, new_cap);
 920         if (xattr_blob)
 921                 ceph_buffer_put(xattr_blob);
 922         return err;
 923 }
 924
 925 /*
 926  * caller should hold session s_mutex.
 927  */
 928 static void update_dentry_lease(struct dentry *dentry,
 929                                 struct ceph_mds_reply_lease *lease,
 930                                 struct ceph_mds_session *session,
 931                                 unsigned long from_time)
 932 {
 933         struct ceph_dentry_info *di = ceph_dentry(dentry);
 934         long unsigned duration = le32_to_cpu(lease->duration_ms);
 935         long unsigned ttl = from_time + (duration * HZ) / 1000;
 936         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 937         struct inode *dir;
 938
 939         /* only track leases on regular dentries */
 940         if (dentry->d_op != &ceph_dentry_ops)
 941                 return;
 942
 943         spin_lock(&dentry->d_lock);
 944         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 945              dentry, duration, ttl);
 946
 947         /* make lease_rdcache_gen match directory */
 948         dir = d_inode(dentry->d_parent);
 949         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 950
 951         if (duration == 0)
 952                 goto out_unlock;
 953
 954         if (di->lease_gen == session->s_cap_gen &&
 955             time_before(ttl, dentry->d_time))
 956                 goto out_unlock;  /* we already have a newer lease. */
 957
 958         if (di->lease_session && di->lease_session != session)
 959                 goto out_unlock;
 960
 961         ceph_dentry_lru_touch(dentry);
 962
 963         if (!di->lease_session)
 964                 di->lease_session = ceph_get_mds_session(session);
 965         di->lease_gen = session->s_cap_gen;
 966         di->lease_seq = le32_to_cpu(lease->seq);
 967         di->lease_renew_after = half_ttl;
 968         di->lease_renew_from = 0;
 969         dentry->d_time = ttl;
 970 out_unlock:
 971         spin_unlock(&dentry->d_lock);
 972         return;
 973 }
 974
 975 /*
 976  * splice a dentry to an inode.
 977  * caller must hold directory i_mutex for this to be safe.
 978  *
 979  * we will only rehash the resulting dentry if @prehash is
 980  * true; @prehash will be set to false (for the benefit of
 981  * the caller) if we fail.
 982  */
 983 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 984                                     bool *prehash)
 985 {
 986         struct dentry *realdn;
 987
 988         BUG_ON(d_inode(dn));
 989
 990         /* dn must be unhashed */
 991         if (!d_unhashed(dn))
 992                 d_drop(dn);
 993         realdn = d_splice_alias(in, dn);
 994         if (IS_ERR(realdn)) {
 995                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 996                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 997                 if (prehash)
 998                         *prehash = false; /* don't rehash on error */
 999                 dn = realdn; /* note realdn contains the error */
1000                 goto out;
1001         } else if (realdn) {
1002                 dout("dn %p (%d) spliced with %p (%d) "
1003                      "inode %p ino %llx.%llx\n",
1004                      dn, d_count(dn),
1005                      realdn, d_count(realdn),
1006                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1007                 dput(dn);
1008                 dn = realdn;
1009         } else {
1010                 BUG_ON(!ceph_dentry(dn));
1011                 dout("dn %p attached to %p ino %llx.%llx\n",
1012                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1013         }
1014         if ((!prehash || *prehash) && d_unhashed(dn))
1015                 d_rehash(dn);
1016 out:
1017         return dn;
1018 }
1019
1020 /*
1021  * Incorporate results into the local cache.  This is either just
1022  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1023  * after a lookup).
1024  *
1025  * A reply may contain
1026  *         a directory inode along with a dentry.
1027  *  and/or a target inode
1028  *
1029  * Called with snap_rwsem (read).
1030  */
1031 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1032                     struct ceph_mds_session *session)
1033 {
1034         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1035         struct inode *in = NULL;
1036         struct ceph_vino vino;
1037         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1038         int err = 0;
1039
1040         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1041              rinfo->head->is_dentry, rinfo->head->is_target);
1042
1043 #if 0
1044         /*
1045          * Debugging hook:
1046          *
1047          * If we resend completed ops to a recovering mds, we get no
1048          * trace.  Since that is very rare, pretend this is the case
1049          * to ensure the 'no trace' handlers in the callers behave.
1050          *
1051          * Fill in inodes unconditionally to avoid breaking cap
1052          * invariants.
1053          */
1054         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1055                 pr_info("fill_trace faking empty trace on %lld %s\n",
1056                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1057                 if (rinfo->head->is_dentry) {
1058                         rinfo->head->is_dentry = 0;
1059                         err = fill_inode(req->r_locked_dir,
1060                                          &rinfo->diri, rinfo->dirfrag,
1061                                          session, req->r_request_started, -1);
1062                 }
1063                 if (rinfo->head->is_target) {
1064                         rinfo->head->is_target = 0;
1065                         ininfo = rinfo->targeti.in;
1066                         vino.ino = le64_to_cpu(ininfo->ino);
1067                         vino.snap = le64_to_cpu(ininfo->snapid);
1068                         in = ceph_get_inode(sb, vino);
1069                         err = fill_inode(in, &rinfo->targeti, NULL,
1070                                          session, req->r_request_started,
1071                                          req->r_fmode);
1072                         iput(in);
1073                 }
1074         }
1075 #endif
1076
1077         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1078                 dout("fill_trace reply is empty!\n");
1079                 if (rinfo->head->result == 0 && req->r_locked_dir)
1080                         ceph_invalidate_dir_request(req);
1081                 return 0;
1082         }
1083
1084         if (rinfo->head->is_dentry) {
1085                 struct inode *dir = req->r_locked_dir;
1086
1087                 if (dir) {
1088                         err = fill_inode(dir, NULL,
1089                                          &rinfo->diri, rinfo->dirfrag,
1090                                          session, req->r_request_started, -1,
1091                                          &req->r_caps_reservation);
1092                         if (err < 0)
1093                                 goto done;
1094                 } else {
1095                         WARN_ON_ONCE(1);
1096                 }
1097
1098                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1099                         struct qstr dname;
1100                         struct dentry *dn, *parent;
1101
1102                         BUG_ON(!rinfo->head->is_target);
1103                         BUG_ON(req->r_dentry);
1104
1105                         parent = d_find_any_alias(dir);
1106                         BUG_ON(!parent);
1107
1108                         dname.name = rinfo->dname;
1109                         dname.len = rinfo->dname_len;
1110                         dname.hash = full_name_hash(dname.name, dname.len);
1111                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1112                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1113 retry_lookup:
1114                         dn = d_lookup(parent, &dname);
1115                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1116                              parent, dname.len, dname.name, dn);
1117
1118                         if (!dn) {
1119                                 dn = d_alloc(parent, &dname);
1120                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1121                                      dname.len, dname.name, dn);
1122                                 if (dn == NULL) {
1123                                         dput(parent);
1124                                         err = -ENOMEM;
1125                                         goto done;
1126                                 }
1127                                 err = ceph_init_dentry(dn);
1128                                 if (err < 0) {
1129                                         dput(dn);
1130                                         dput(parent);
1131                                         goto done;
1132                                 }
1133                         } else if (d_really_is_positive(dn) &&
1134                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1135                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1136                                 dout(" dn %p points to wrong inode %p\n",
1137                                      dn, d_inode(dn));
1138                                 d_delete(dn);
1139                                 dput(dn);
1140                                 goto retry_lookup;
1141                         }
1142
1143                         req->r_dentry = dn;
1144                         dput(parent);
1145                 }
1146         }
1147
1148         if (rinfo->head->is_target) {
1149                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1150                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1151
1152                 in = ceph_get_inode(sb, vino);
1153                 if (IS_ERR(in)) {
1154                         err = PTR_ERR(in);
1155                         goto done;
1156                 }
1157                 req->r_target_inode = in;
1158
1159                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1160                                 session, req->r_request_started,
1161                                 (!req->r_aborted && rinfo->head->result == 0) ?
1162                                 req->r_fmode : -1,
1163                                 &req->r_caps_reservation);
1164                 if (err < 0) {
1165                         pr_err("fill_inode badness %p %llx.%llx\n",
1166                                 in, ceph_vinop(in));
1167                         goto done;
1168                 }
1169         }
1170
1171         /*
1172          * ignore null lease/binding on snapdir ENOENT, or else we
1173          * will have trouble splicing in the virtual snapdir later
1174          */
1175         if (rinfo->head->is_dentry && !req->r_aborted &&
1176             req->r_locked_dir &&
1177             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1178                                                fsc->mount_options->snapdir_name,
1179                                                req->r_dentry->d_name.len))) {
1180                 /*
1181                  * lookup link rename   : null -> possibly existing inode
1182                  * mknod symlink mkdir  : null -> new inode
1183                  * unlink               : linked -> null
1184                  */
1185                 struct inode *dir = req->r_locked_dir;
1186                 struct dentry *dn = req->r_dentry;
1187                 bool have_dir_cap, have_lease;
1188
1189                 BUG_ON(!dn);
1190                 BUG_ON(!dir);
1191                 BUG_ON(d_inode(dn->d_parent) != dir);
1192                 BUG_ON(ceph_ino(dir) !=
1193                        le64_to_cpu(rinfo->diri.in->ino));
1194                 BUG_ON(ceph_snap(dir) !=
1195                        le64_to_cpu(rinfo->diri.in->snapid));
1196
1197                 /* do we have a lease on the whole dir? */
1198                 have_dir_cap =
1199                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1200                          CEPH_CAP_FILE_SHARED);
1201
1202                 /* do we have a dn lease? */
1203                 have_lease = have_dir_cap ||
1204                         le32_to_cpu(rinfo->dlease->duration_ms);
1205                 if (!have_lease)
1206                         dout("fill_trace  no dentry lease or dir cap\n");
1207
1208                 /* rename? */
1209                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1210                         struct inode *olddir = req->r_old_dentry_dir;
1211                         BUG_ON(!olddir);
1212
1213                         dout(" src %p '%pd' dst %p '%pd'\n",
1214                              req->r_old_dentry,
1215                              req->r_old_dentry,
1216                              dn, dn);
1217                         dout("fill_trace doing d_move %p -> %p\n",
1218                              req->r_old_dentry, dn);
1219
1220                         /* d_move screws up sibling dentries' offsets */
1221                         ceph_dir_clear_ordered(dir);
1222                         ceph_dir_clear_ordered(olddir);
1223
1224                         d_move(req->r_old_dentry, dn);
1225                         dout(" src %p '%pd' dst %p '%pd'\n",
1226                              req->r_old_dentry,
1227                              req->r_old_dentry,
1228                              dn, dn);
1229
1230                         /* ensure target dentry is invalidated, despite
1231                            rehashing bug in vfs_rename_dir */
1232                         ceph_invalidate_dentry_lease(dn);
1233
1234                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1235                              ceph_dentry(req->r_old_dentry)->offset);
1236
1237                         dn = req->r_old_dentry;  /* use old_dentry */
1238                 }
1239
1240                 /* null dentry? */
1241                 if (!rinfo->head->is_target) {
1242                         dout("fill_trace null dentry\n");
1243                         if (d_really_is_positive(dn)) {
1244                                 ceph_dir_clear_ordered(dir);
1245                                 dout("d_delete %p\n", dn);
1246                                 d_delete(dn);
1247                         } else {
1248                                 dout("d_instantiate %p NULL\n", dn);
1249                                 d_instantiate(dn, NULL);
1250                                 if (have_lease && d_unhashed(dn))
1251                                         d_rehash(dn);
1252                                 update_dentry_lease(dn, rinfo->dlease,
1253                                                     session,
1254                                                     req->r_request_started);
1255                         }
1256                         goto done;
1257                 }
1258
1259                 /* attach proper inode */
1260                 if (d_really_is_negative(dn)) {
1261                         ceph_dir_clear_ordered(dir);
1262                         ihold(in);
1263                         dn = splice_dentry(dn, in, &have_lease);
1264                         if (IS_ERR(dn)) {
1265                                 err = PTR_ERR(dn);
1266                                 goto done;
1267                         }
1268                         req->r_dentry = dn;  /* may have spliced */
1269                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1270                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1271                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1272                              ceph_vinop(in));
1273                         have_lease = false;
1274                 }
1275
1276                 if (have_lease)
1277                         update_dentry_lease(dn, rinfo->dlease, session,
1278                                             req->r_request_started);
1279                 dout(" final dn %p\n", dn);
1280         } else if (!req->r_aborted &&
1281                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1282                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1283                 struct dentry *dn = req->r_dentry;
1284                 struct inode *dir = req->r_locked_dir;
1285
1286                 /* fill out a snapdir LOOKUPSNAP dentry */
1287                 BUG_ON(!dn);
1288                 BUG_ON(!dir);
1289                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1290                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1291                 ceph_dir_clear_ordered(dir);
1292                 ihold(in);
1293                 dn = splice_dentry(dn, in, NULL);
1294                 if (IS_ERR(dn)) {
1295                         err = PTR_ERR(dn);
1296                         goto done;
1297                 }
1298                 req->r_dentry = dn;  /* may have spliced */
1299         }
1300 done:
1301         dout("fill_trace done err=%d\n", err);
1302         return err;
1303 }
1304
1305 /*
1306  * Prepopulate our cache with readdir results, leases, etc.
1307  */
1308 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1309                                            struct ceph_mds_session *session)
1310 {
1311         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1312         int i, err = 0;
1313
1314         for (i = 0; i < rinfo->dir_nr; i++) {
1315                 struct ceph_vino vino;
1316                 struct inode *in;
1317                 int rc;
1318
1319                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1320                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1321
1322                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1323                 if (IS_ERR(in)) {
1324                         err = PTR_ERR(in);
1325                         dout("new_inode badness got %d\n", err);
1326                         continue;
1327                 }
1328                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1329                                 req->r_request_started, -1,
1330                                 &req->r_caps_reservation);
1331                 if (rc < 0) {
1332                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1333                         err = rc;
1334                         continue;
1335                 }
1336         }
1337
1338         return err;
1339 }
1340
1341 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1342 {
1343         if (ctl->page) {
1344                 kunmap(ctl->page);
1345                 page_cache_release(ctl->page);
1346                 ctl->page = NULL;
1347         }
1348 }
1349
1350 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1351                               struct ceph_readdir_cache_control *ctl,
1352                               struct ceph_mds_request *req)
1353 {
1354         struct ceph_inode_info *ci = ceph_inode(dir);
1355         unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry*);
1356         unsigned idx = ctl->index % nsize;
1357         pgoff_t pgoff = ctl->index / nsize;
1358
1359         if (!ctl->page || pgoff != page_index(ctl->page)) {
1360                 ceph_readdir_cache_release(ctl);
1361                 ctl->page  = grab_cache_page(&dir->i_data, pgoff);
1362                 if (!ctl->page) {
1363                         ctl->index = -1;
1364                         return -ENOMEM;
1365                 }
1366                 /* reading/filling the cache are serialized by
1367                  * i_mutex, no need to use page lock */
1368                 unlock_page(ctl->page);
1369                 ctl->dentries = kmap(ctl->page);
1370         }
1371
1372         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1373             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1374                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1375                 ctl->dentries[idx] = dn;
1376                 ctl->index++;
1377         } else {
1378                 dout("disable readdir cache\n");
1379                 ctl->index = -1;
1380         }
1381         return 0;
1382 }
1383
1384 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1385                              struct ceph_mds_session *session)
1386 {
1387         struct dentry *parent = req->r_dentry;
1388         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1389         struct qstr dname;
1390         struct dentry *dn;
1391         struct inode *in;
1392         int err = 0, ret, i;
1393         struct inode *snapdir = NULL;
1394         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1395         struct ceph_dentry_info *di;
1396         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1397         struct ceph_readdir_cache_control cache_ctl = {};
1398
1399         if (req->r_aborted)
1400                 return readdir_prepopulate_inodes_only(req, session);
1401
1402         if (rinfo->dir_dir &&
1403             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1404                 dout("readdir_prepopulate got new frag %x -> %x\n",
1405                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1406                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1407                 if (ceph_frag_is_leftmost(frag))
1408                         req->r_readdir_offset = 2;
1409                 else
1410                         req->r_readdir_offset = 0;
1411         }
1412
1413         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1414                 snapdir = ceph_get_snapdir(d_inode(parent));
1415                 parent = d_find_alias(snapdir);
1416                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1417                      rinfo->dir_nr, parent);
1418         } else {
1419                 dout("readdir_prepopulate %d items under dn %p\n",
1420                      rinfo->dir_nr, parent);
1421                 if (rinfo->dir_dir)
1422                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1423         }
1424
1425         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1426                 /* note dir version at start of readdir so we can tell
1427                  * if any dentries get dropped */
1428                 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1429                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1430                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1431                 req->r_readdir_cache_idx = 0;
1432         }
1433
1434         cache_ctl.index = req->r_readdir_cache_idx;
1435
1436         /* FIXME: release caps/leases if error occurs */
1437         for (i = 0; i < rinfo->dir_nr; i++) {
1438                 struct ceph_vino vino;
1439
1440                 dname.name = rinfo->dir_dname[i];
1441                 dname.len = rinfo->dir_dname_len[i];
1442                 dname.hash = full_name_hash(dname.name, dname.len);
1443
1444                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1445                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1446
1447 retry_lookup:
1448                 dn = d_lookup(parent, &dname);
1449                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1450                      parent, dname.len, dname.name, dn);
1451
1452                 if (!dn) {
1453                         dn = d_alloc(parent, &dname);
1454                         dout("d_alloc %p '%.*s' = %p\n", parent,
1455                              dname.len, dname.name, dn);
1456                         if (dn == NULL) {
1457                                 dout("d_alloc badness\n");
1458                                 err = -ENOMEM;
1459                                 goto out;
1460                         }
1461                         ret = ceph_init_dentry(dn);
1462                         if (ret < 0) {
1463                                 dput(dn);
1464                                 err = ret;
1465                                 goto out;
1466                         }
1467                 } else if (d_really_is_positive(dn) &&
1468                            (ceph_ino(d_inode(dn)) != vino.ino ||
1469                             ceph_snap(d_inode(dn)) != vino.snap)) {
1470                         dout(" dn %p points to wrong inode %p\n",
1471                              dn, d_inode(dn));
1472                         d_delete(dn);
1473                         dput(dn);
1474                         goto retry_lookup;
1475                 }
1476
1477                 /* inode */
1478                 if (d_really_is_positive(dn)) {
1479                         in = d_inode(dn);
1480                 } else {
1481                         in = ceph_get_inode(parent->d_sb, vino);
1482                         if (IS_ERR(in)) {
1483                                 dout("new_inode badness\n");
1484                                 d_drop(dn);
1485                                 dput(dn);
1486                                 err = PTR_ERR(in);
1487                                 goto out;
1488                         }
1489                 }
1490
1491                 ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1492                                  req->r_request_started, -1,
1493                                  &req->r_caps_reservation);
1494                 if (ret < 0) {
1495                         pr_err("fill_inode badness on %p\n", in);
1496                         if (d_really_is_negative(dn))
1497                                 iput(in);
1498                         d_drop(dn);
1499                         err = ret;
1500                         goto next_item;
1501                 }
1502
1503                 if (d_really_is_negative(dn)) {
1504                         struct dentry *realdn = splice_dentry(dn, in, NULL);
1505                         if (IS_ERR(realdn)) {
1506                                 err = PTR_ERR(realdn);
1507                                 d_drop(dn);
1508                                 dn = NULL;
1509                                 goto next_item;
1510                         }
1511                         dn = realdn;
1512                 }
1513
1514                 di = dn->d_fsdata;
1515                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1516
1517                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1518                                     req->r_session,
1519                                     req->r_request_started);
1520
1521                 if (err == 0 && cache_ctl.index >= 0) {
1522                         ret = fill_readdir_cache(d_inode(parent), dn,
1523                                                  &cache_ctl, req);
1524                         if (ret < 0)
1525                                 err = ret;
1526                 }
1527 next_item:
1528                 if (dn)
1529                         dput(dn);
1530         }
1531 out:
1532         if (err == 0) {
1533                 req->r_did_prepopulate = true;
1534                 req->r_readdir_cache_idx = cache_ctl.index;
1535         }
1536         ceph_readdir_cache_release(&cache_ctl);
1537         if (snapdir) {
1538                 iput(snapdir);
1539                 dput(parent);
1540         }
1541         dout("readdir_prepopulate done\n");
1542         return err;
1543 }
1544
1545 int ceph_inode_set_size(struct inode *inode, loff_t size)
1546 {
1547         struct ceph_inode_info *ci = ceph_inode(inode);
1548         int ret = 0;
1549
1550         spin_lock(&ci->i_ceph_lock);
1551         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1552         i_size_write(inode, size);
1553         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1554
1555         /* tell the MDS if we are approaching max_size */
1556         if ((size << 1) >= ci->i_max_size &&
1557             (ci->i_reported_size << 1) < ci->i_max_size)
1558                 ret = 1;
1559
1560         spin_unlock(&ci->i_ceph_lock);
1561         return ret;
1562 }
1563
1564 /*
1565  * Write back inode data in a worker thread.  (This can't be done
1566  * in the message handler context.)
1567  */
1568 void ceph_queue_writeback(struct inode *inode)
1569 {
1570         ihold(inode);
1571         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1572                        &ceph_inode(inode)->i_wb_work)) {
1573                 dout("ceph_queue_writeback %p\n", inode);
1574         } else {
1575                 dout("ceph_queue_writeback %p failed\n", inode);
1576                 iput(inode);
1577         }
1578 }
1579
1580 static void ceph_writeback_work(struct work_struct *work)
1581 {
1582         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1583                                                   i_wb_work);
1584         struct inode *inode = &ci->vfs_inode;
1585
1586         dout("writeback %p\n", inode);
1587         filemap_fdatawrite(&inode->i_data);
1588         iput(inode);
1589 }
1590
1591 /*
1592  * queue an async invalidation
1593  */
1594 void ceph_queue_invalidate(struct inode *inode)
1595 {
1596         ihold(inode);
1597         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1598                        &ceph_inode(inode)->i_pg_inv_work)) {
1599                 dout("ceph_queue_invalidate %p\n", inode);
1600         } else {
1601                 dout("ceph_queue_invalidate %p failed\n", inode);
1602                 iput(inode);
1603         }
1604 }
1605
1606 /*
1607  * Invalidate inode pages in a worker thread.  (This can't be done
1608  * in the message handler context.)
1609  */
1610 static void ceph_invalidate_work(struct work_struct *work)
1611 {
1612         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1613                                                   i_pg_inv_work);
1614         struct inode *inode = &ci->vfs_inode;
1615         u32 orig_gen;
1616         int check = 0;
1617
1618         mutex_lock(&ci->i_truncate_mutex);
1619         spin_lock(&ci->i_ceph_lock);
1620         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1621              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1622         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1623                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1624                         check = 1;
1625                 spin_unlock(&ci->i_ceph_lock);
1626                 mutex_unlock(&ci->i_truncate_mutex);
1627                 goto out;
1628         }
1629         orig_gen = ci->i_rdcache_gen;
1630         spin_unlock(&ci->i_ceph_lock);
1631
1632         truncate_pagecache(inode, 0);
1633
1634         spin_lock(&ci->i_ceph_lock);
1635         if (orig_gen == ci->i_rdcache_gen &&
1636             orig_gen == ci->i_rdcache_revoking) {
1637                 dout("invalidate_pages %p gen %d successful\n", inode,
1638                      ci->i_rdcache_gen);
1639                 ci->i_rdcache_revoking--;
1640                 check = 1;
1641         } else {
1642                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1643                      inode, orig_gen, ci->i_rdcache_gen,
1644                      ci->i_rdcache_revoking);
1645                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1646                         check = 1;
1647         }
1648         spin_unlock(&ci->i_ceph_lock);
1649         mutex_unlock(&ci->i_truncate_mutex);
1650 out:
1651         if (check)
1652                 ceph_check_caps(ci, 0, NULL);
1653         iput(inode);
1654 }
1655
1656
1657 /*
1658  * called by trunc_wq;
1659  *
1660  * We also truncate in a separate thread as well.
1661  */
1662 static void ceph_vmtruncate_work(struct work_struct *work)
1663 {
1664         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1665                                                   i_vmtruncate_work);
1666         struct inode *inode = &ci->vfs_inode;
1667
1668         dout("vmtruncate_work %p\n", inode);
1669         __ceph_do_pending_vmtruncate(inode);
1670         iput(inode);
1671 }
1672
1673 /*
1674  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1675  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1676  */
1677 void ceph_queue_vmtruncate(struct inode *inode)
1678 {
1679         struct ceph_inode_info *ci = ceph_inode(inode);
1680
1681         ihold(inode);
1682
1683         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1684                        &ci->i_vmtruncate_work)) {
1685                 dout("ceph_queue_vmtruncate %p\n", inode);
1686         } else {
1687                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1688                      inode, ci->i_truncate_pending);
1689                 iput(inode);
1690         }
1691 }
1692
1693 /*
1694  * Make sure any pending truncation is applied before doing anything
1695  * that may depend on it.
1696  */
1697 void __ceph_do_pending_vmtruncate(struct inode *inode)
1698 {
1699         struct ceph_inode_info *ci = ceph_inode(inode);
1700         u64 to;
1701         int wrbuffer_refs, finish = 0;
1702
1703         mutex_lock(&ci->i_truncate_mutex);
1704 retry:
1705         spin_lock(&ci->i_ceph_lock);
1706         if (ci->i_truncate_pending == 0) {
1707                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1708                 spin_unlock(&ci->i_ceph_lock);
1709                 mutex_unlock(&ci->i_truncate_mutex);
1710                 return;
1711         }
1712
1713         /*
1714          * make sure any dirty snapped pages are flushed before we
1715          * possibly truncate them.. so write AND block!
1716          */
1717         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1718                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1719                      inode);
1720                 spin_unlock(&ci->i_ceph_lock);
1721                 filemap_write_and_wait_range(&inode->i_data, 0,
1722                                              inode->i_sb->s_maxbytes);
1723                 goto retry;
1724         }
1725
1726         /* there should be no reader or writer */
1727         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1728
1729         to = ci->i_truncate_size;
1730         wrbuffer_refs = ci->i_wrbuffer_ref;
1731         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1732              ci->i_truncate_pending, to);
1733         spin_unlock(&ci->i_ceph_lock);
1734
1735         truncate_pagecache(inode, to);
1736
1737         spin_lock(&ci->i_ceph_lock);
1738         if (to == ci->i_truncate_size) {
1739                 ci->i_truncate_pending = 0;
1740                 finish = 1;
1741         }
1742         spin_unlock(&ci->i_ceph_lock);
1743         if (!finish)
1744                 goto retry;
1745
1746         mutex_unlock(&ci->i_truncate_mutex);
1747
1748         if (wrbuffer_refs == 0)
1749                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1750
1751         wake_up_all(&ci->i_cap_wq);
1752 }
1753
1754 /*
1755  * symlinks
1756  */
1757 static const struct inode_operations ceph_symlink_iops = {
1758         .readlink = generic_readlink,
1759         .get_link = simple_get_link,
1760         .setattr = ceph_setattr,
1761         .getattr = ceph_getattr,
1762         .setxattr = ceph_setxattr,
1763         .getxattr = ceph_getxattr,
1764         .listxattr = ceph_listxattr,
1765         .removexattr = ceph_removexattr,
1766 };
1767
1768 /*
1769  * setattr
1770  */
1771 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1772 {
1773         struct inode *inode = d_inode(dentry);
1774         struct ceph_inode_info *ci = ceph_inode(inode);
1775         const unsigned int ia_valid = attr->ia_valid;
1776         struct ceph_mds_request *req;
1777         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1778         struct ceph_cap_flush *prealloc_cf;
1779         int issued;
1780         int release = 0, dirtied = 0;
1781         int mask = 0;
1782         int err = 0;
1783         int inode_dirty_flags = 0;
1784         bool lock_snap_rwsem = false;
1785
1786         if (ceph_snap(inode) != CEPH_NOSNAP)
1787                 return -EROFS;
1788
1789         err = inode_change_ok(inode, attr);
1790         if (err != 0)
1791                 return err;
1792
1793         prealloc_cf = ceph_alloc_cap_flush();
1794         if (!prealloc_cf)
1795                 return -ENOMEM;
1796
1797         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1798                                        USE_AUTH_MDS);
1799         if (IS_ERR(req)) {
1800                 ceph_free_cap_flush(prealloc_cf);
1801                 return PTR_ERR(req);
1802         }
1803
1804         spin_lock(&ci->i_ceph_lock);
1805         issued = __ceph_caps_issued(ci, NULL);
1806
1807         if (!ci->i_head_snapc &&
1808             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1809                 lock_snap_rwsem = true;
1810                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1811                         spin_unlock(&ci->i_ceph_lock);
1812                         down_read(&mdsc->snap_rwsem);
1813                         spin_lock(&ci->i_ceph_lock);
1814                         issued = __ceph_caps_issued(ci, NULL);
1815                 }
1816         }
1817
1818         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1819
1820         if (ia_valid & ATTR_UID) {
1821                 dout("setattr %p uid %d -> %d\n", inode,
1822                      from_kuid(&init_user_ns, inode->i_uid),
1823                      from_kuid(&init_user_ns, attr->ia_uid));
1824                 if (issued & CEPH_CAP_AUTH_EXCL) {
1825                         inode->i_uid = attr->ia_uid;
1826                         dirtied |= CEPH_CAP_AUTH_EXCL;
1827                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1828                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1829                         req->r_args.setattr.uid = cpu_to_le32(
1830                                 from_kuid(&init_user_ns, attr->ia_uid));
1831                         mask |= CEPH_SETATTR_UID;
1832                         release |= CEPH_CAP_AUTH_SHARED;
1833                 }
1834         }
1835         if (ia_valid & ATTR_GID) {
1836                 dout("setattr %p gid %d -> %d\n", inode,
1837                      from_kgid(&init_user_ns, inode->i_gid),
1838                      from_kgid(&init_user_ns, attr->ia_gid));
1839                 if (issued & CEPH_CAP_AUTH_EXCL) {
1840                         inode->i_gid = attr->ia_gid;
1841                         dirtied |= CEPH_CAP_AUTH_EXCL;
1842                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1843                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1844                         req->r_args.setattr.gid = cpu_to_le32(
1845                                 from_kgid(&init_user_ns, attr->ia_gid));
1846                         mask |= CEPH_SETATTR_GID;
1847                         release |= CEPH_CAP_AUTH_SHARED;
1848                 }
1849         }
1850         if (ia_valid & ATTR_MODE) {
1851                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1852                      attr->ia_mode);
1853                 if (issued & CEPH_CAP_AUTH_EXCL) {
1854                         inode->i_mode = attr->ia_mode;
1855                         dirtied |= CEPH_CAP_AUTH_EXCL;
1856                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1857                            attr->ia_mode != inode->i_mode) {
1858                         inode->i_mode = attr->ia_mode;
1859                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1860                         mask |= CEPH_SETATTR_MODE;
1861                         release |= CEPH_CAP_AUTH_SHARED;
1862                 }
1863         }
1864
1865         if (ia_valid & ATTR_ATIME) {
1866                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1867                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1868                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1869                 if (issued & CEPH_CAP_FILE_EXCL) {
1870                         ci->i_time_warp_seq++;
1871                         inode->i_atime = attr->ia_atime;
1872                         dirtied |= CEPH_CAP_FILE_EXCL;
1873                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1874                            timespec_compare(&inode->i_atime,
1875                                             &attr->ia_atime) < 0) {
1876                         inode->i_atime = attr->ia_atime;
1877                         dirtied |= CEPH_CAP_FILE_WR;
1878                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1879                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1880                         ceph_encode_timespec(&req->r_args.setattr.atime,
1881                                              &attr->ia_atime);
1882                         mask |= CEPH_SETATTR_ATIME;
1883                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1884                                 CEPH_CAP_FILE_WR;
1885                 }
1886         }
1887         if (ia_valid & ATTR_MTIME) {
1888                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1889                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1890                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1891                 if (issued & CEPH_CAP_FILE_EXCL) {
1892                         ci->i_time_warp_seq++;
1893                         inode->i_mtime = attr->ia_mtime;
1894                         dirtied |= CEPH_CAP_FILE_EXCL;
1895                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1896                            timespec_compare(&inode->i_mtime,
1897                                             &attr->ia_mtime) < 0) {
1898                         inode->i_mtime = attr->ia_mtime;
1899                         dirtied |= CEPH_CAP_FILE_WR;
1900                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1901                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1902                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1903                                              &attr->ia_mtime);
1904                         mask |= CEPH_SETATTR_MTIME;
1905                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1906                                 CEPH_CAP_FILE_WR;
1907                 }
1908         }
1909         if (ia_valid & ATTR_SIZE) {
1910                 dout("setattr %p size %lld -> %lld\n", inode,
1911                      inode->i_size, attr->ia_size);
1912                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1913                     attr->ia_size > inode->i_size) {
1914                         i_size_write(inode, attr->ia_size);
1915                         inode->i_blocks =
1916                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1917                         inode->i_ctime = attr->ia_ctime;
1918                         ci->i_reported_size = attr->ia_size;
1919                         dirtied |= CEPH_CAP_FILE_EXCL;
1920                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1921                            attr->ia_size != inode->i_size) {
1922                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1923                         req->r_args.setattr.old_size =
1924                                 cpu_to_le64(inode->i_size);
1925                         mask |= CEPH_SETATTR_SIZE;
1926                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1927                                 CEPH_CAP_FILE_WR;
1928                 }
1929         }
1930
1931         /* these do nothing */
1932         if (ia_valid & ATTR_CTIME) {
1933                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1934                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1935                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1936                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1937                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1938                      only ? "ctime only" : "ignored");
1939                 inode->i_ctime = attr->ia_ctime;
1940                 if (only) {
1941                         /*
1942                          * if kernel wants to dirty ctime but nothing else,
1943                          * we need to choose a cap to dirty under, or do
1944                          * a almost-no-op setattr
1945                          */
1946                         if (issued & CEPH_CAP_AUTH_EXCL)
1947                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1948                         else if (issued & CEPH_CAP_FILE_EXCL)
1949                                 dirtied |= CEPH_CAP_FILE_EXCL;
1950                         else if (issued & CEPH_CAP_XATTR_EXCL)
1951                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1952                         else
1953                                 mask |= CEPH_SETATTR_CTIME;
1954                 }
1955         }
1956         if (ia_valid & ATTR_FILE)
1957                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1958
1959         if (dirtied) {
1960                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
1961                                                            &prealloc_cf);
1962                 inode->i_ctime = CURRENT_TIME;
1963         }
1964
1965         release &= issued;
1966         spin_unlock(&ci->i_ceph_lock);
1967         if (lock_snap_rwsem)
1968                 up_read(&mdsc->snap_rwsem);
1969
1970         if (inode_dirty_flags)
1971                 __mark_inode_dirty(inode, inode_dirty_flags);
1972
1973         if (ia_valid & ATTR_MODE) {
1974                 err = posix_acl_chmod(inode, attr->ia_mode);
1975                 if (err)
1976                         goto out_put;
1977         }
1978
1979         if (mask) {
1980                 req->r_inode = inode;
1981                 ihold(inode);
1982                 req->r_inode_drop = release;
1983                 req->r_args.setattr.mask = cpu_to_le32(mask);
1984                 req->r_num_caps = 1;
1985                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1986         }
1987         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1988              ceph_cap_string(dirtied), mask);
1989
1990         ceph_mdsc_put_request(req);
1991         if (mask & CEPH_SETATTR_SIZE)
1992                 __ceph_do_pending_vmtruncate(inode);
1993         ceph_free_cap_flush(prealloc_cf);
1994         return err;
1995 out_put:
1996         ceph_mdsc_put_request(req);
1997         ceph_free_cap_flush(prealloc_cf);
1998         return err;
1999 }
2000
2001 /*
2002  * Verify that we have a lease on the given mask.  If not,
2003  * do a getattr against an mds.
2004  */
2005 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2006                       int mask, bool force)
2007 {
2008         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2009         struct ceph_mds_client *mdsc = fsc->mdsc;
2010         struct ceph_mds_request *req;
2011         int err;
2012
2013         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2014                 dout("do_getattr inode %p SNAPDIR\n", inode);
2015                 return 0;
2016         }
2017
2018         dout("do_getattr inode %p mask %s mode 0%o\n",
2019              inode, ceph_cap_string(mask), inode->i_mode);
2020         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2021                 return 0;
2022
2023         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2024         if (IS_ERR(req))
2025                 return PTR_ERR(req);
2026         req->r_inode = inode;
2027         ihold(inode);
2028         req->r_num_caps = 1;
2029         req->r_args.getattr.mask = cpu_to_le32(mask);
2030         req->r_locked_page = locked_page;
2031         err = ceph_mdsc_do_request(mdsc, NULL, req);
2032         if (locked_page && err == 0) {
2033                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2034                 if (inline_version == 0) {
2035                         /* the reply is supposed to contain inline data */
2036                         err = -EINVAL;
2037                 } else if (inline_version == CEPH_INLINE_NONE) {
2038                         err = -ENODATA;
2039                 } else {
2040                         err = req->r_reply_info.targeti.inline_len;
2041                 }
2042         }
2043         ceph_mdsc_put_request(req);
2044         dout("do_getattr result=%d\n", err);
2045         return err;
2046 }
2047
2048
2049 /*
2050  * Check inode permissions.  We verify we have a valid value for
2051  * the AUTH cap, then call the generic handler.
2052  */
2053 int ceph_permission(struct inode *inode, int mask)
2054 {
2055         int err;
2056
2057         if (mask & MAY_NOT_BLOCK)
2058                 return -ECHILD;
2059
2060         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2061
2062         if (!err)
2063                 err = generic_permission(inode, mask);
2064         return err;
2065 }
2066
2067 /*
2068  * Get all attributes.  Hopefully somedata we'll have a statlite()
2069  * and can limit the fields we require to be accurate.
2070  */
2071 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2072                  struct kstat *stat)
2073 {
2074         struct inode *inode = d_inode(dentry);
2075         struct ceph_inode_info *ci = ceph_inode(inode);
2076         int err;
2077
2078         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2079         if (!err) {
2080                 generic_fillattr(inode, stat);
2081                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2082                 if (ceph_snap(inode) != CEPH_NOSNAP)
2083                         stat->dev = ceph_snap(inode);
2084                 else
2085                         stat->dev = 0;
2086                 if (S_ISDIR(inode->i_mode)) {
2087                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2088                                                 RBYTES))
2089                                 stat->size = ci->i_rbytes;
2090                         else
2091                                 stat->size = ci->i_files + ci->i_subdirs;
2092                         stat->blocks = 0;
2093                         stat->blksize = 65536;
2094                 }
2095         }
2096         return err;
2097 }