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 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
 980 {
 981         struct dentry *realdn;
 982
 983         BUG_ON(d_inode(dn));
 984
 985         /* dn must be unhashed */
 986         if (!d_unhashed(dn))
 987                 d_drop(dn);
 988         realdn = d_splice_alias(in, dn);
 989         if (IS_ERR(realdn)) {
 990                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 991                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 992                 dn = realdn; /* note realdn contains the error */
 993                 goto out;
 994         } else if (realdn) {
 995                 dout("dn %p (%d) spliced with %p (%d) "
 996                      "inode %p ino %llx.%llx\n",
 997                      dn, d_count(dn),
 998                      realdn, d_count(realdn),
 999                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1000                 dput(dn);
1001                 dn = realdn;
1002         } else {
1003                 BUG_ON(!ceph_dentry(dn));
1004                 dout("dn %p attached to %p ino %llx.%llx\n",
1005                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1006         }
1007 out:
1008         return dn;
1009 }
1010
1011 /*
1012  * Incorporate results into the local cache.  This is either just
1013  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1014  * after a lookup).
1015  *
1016  * A reply may contain
1017  *         a directory inode along with a dentry.
1018  *  and/or a target inode
1019  *
1020  * Called with snap_rwsem (read).
1021  */
1022 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1023                     struct ceph_mds_session *session)
1024 {
1025         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1026         struct inode *in = NULL;
1027         struct ceph_vino vino;
1028         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1029         int err = 0;
1030
1031         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1032              rinfo->head->is_dentry, rinfo->head->is_target);
1033
1034 #if 0
1035         /*
1036          * Debugging hook:
1037          *
1038          * If we resend completed ops to a recovering mds, we get no
1039          * trace.  Since that is very rare, pretend this is the case
1040          * to ensure the 'no trace' handlers in the callers behave.
1041          *
1042          * Fill in inodes unconditionally to avoid breaking cap
1043          * invariants.
1044          */
1045         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1046                 pr_info("fill_trace faking empty trace on %lld %s\n",
1047                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1048                 if (rinfo->head->is_dentry) {
1049                         rinfo->head->is_dentry = 0;
1050                         err = fill_inode(req->r_locked_dir,
1051                                          &rinfo->diri, rinfo->dirfrag,
1052                                          session, req->r_request_started, -1);
1053                 }
1054                 if (rinfo->head->is_target) {
1055                         rinfo->head->is_target = 0;
1056                         ininfo = rinfo->targeti.in;
1057                         vino.ino = le64_to_cpu(ininfo->ino);
1058                         vino.snap = le64_to_cpu(ininfo->snapid);
1059                         in = ceph_get_inode(sb, vino);
1060                         err = fill_inode(in, &rinfo->targeti, NULL,
1061                                          session, req->r_request_started,
1062                                          req->r_fmode);
1063                         iput(in);
1064                 }
1065         }
1066 #endif
1067
1068         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1069                 dout("fill_trace reply is empty!\n");
1070                 if (rinfo->head->result == 0 && req->r_locked_dir)
1071                         ceph_invalidate_dir_request(req);
1072                 return 0;
1073         }
1074
1075         if (rinfo->head->is_dentry) {
1076                 struct inode *dir = req->r_locked_dir;
1077
1078                 if (dir) {
1079                         err = fill_inode(dir, NULL,
1080                                          &rinfo->diri, rinfo->dirfrag,
1081                                          session, req->r_request_started, -1,
1082                                          &req->r_caps_reservation);
1083                         if (err < 0)
1084                                 goto done;
1085                 } else {
1086                         WARN_ON_ONCE(1);
1087                 }
1088
1089                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1090                         struct qstr dname;
1091                         struct dentry *dn, *parent;
1092
1093                         BUG_ON(!rinfo->head->is_target);
1094                         BUG_ON(req->r_dentry);
1095
1096                         parent = d_find_any_alias(dir);
1097                         BUG_ON(!parent);
1098
1099                         dname.name = rinfo->dname;
1100                         dname.len = rinfo->dname_len;
1101                         dname.hash = full_name_hash(dname.name, dname.len);
1102                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1103                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1104 retry_lookup:
1105                         dn = d_lookup(parent, &dname);
1106                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1107                              parent, dname.len, dname.name, dn);
1108
1109                         if (!dn) {
1110                                 dn = d_alloc(parent, &dname);
1111                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1112                                      dname.len, dname.name, dn);
1113                                 if (dn == NULL) {
1114                                         dput(parent);
1115                                         err = -ENOMEM;
1116                                         goto done;
1117                                 }
1118                                 err = ceph_init_dentry(dn);
1119                                 if (err < 0) {
1120                                         dput(dn);
1121                                         dput(parent);
1122                                         goto done;
1123                                 }
1124                         } else if (d_really_is_positive(dn) &&
1125                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1126                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1127                                 dout(" dn %p points to wrong inode %p\n",
1128                                      dn, d_inode(dn));
1129                                 d_delete(dn);
1130                                 dput(dn);
1131                                 goto retry_lookup;
1132                         }
1133
1134                         req->r_dentry = dn;
1135                         dput(parent);
1136                 }
1137         }
1138
1139         if (rinfo->head->is_target) {
1140                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1141                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1142
1143                 in = ceph_get_inode(sb, vino);
1144                 if (IS_ERR(in)) {
1145                         err = PTR_ERR(in);
1146                         goto done;
1147                 }
1148                 req->r_target_inode = in;
1149
1150                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1151                                 session, req->r_request_started,
1152                                 (!req->r_aborted && rinfo->head->result == 0) ?
1153                                 req->r_fmode : -1,
1154                                 &req->r_caps_reservation);
1155                 if (err < 0) {
1156                         pr_err("fill_inode badness %p %llx.%llx\n",
1157                                 in, ceph_vinop(in));
1158                         goto done;
1159                 }
1160         }
1161
1162         /*
1163          * ignore null lease/binding on snapdir ENOENT, or else we
1164          * will have trouble splicing in the virtual snapdir later
1165          */
1166         if (rinfo->head->is_dentry && !req->r_aborted &&
1167             req->r_locked_dir &&
1168             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1169                                                fsc->mount_options->snapdir_name,
1170                                                req->r_dentry->d_name.len))) {
1171                 /*
1172                  * lookup link rename   : null -> possibly existing inode
1173                  * mknod symlink mkdir  : null -> new inode
1174                  * unlink               : linked -> null
1175                  */
1176                 struct inode *dir = req->r_locked_dir;
1177                 struct dentry *dn = req->r_dentry;
1178                 bool have_dir_cap, have_lease;
1179
1180                 BUG_ON(!dn);
1181                 BUG_ON(!dir);
1182                 BUG_ON(d_inode(dn->d_parent) != dir);
1183                 BUG_ON(ceph_ino(dir) !=
1184                        le64_to_cpu(rinfo->diri.in->ino));
1185                 BUG_ON(ceph_snap(dir) !=
1186                        le64_to_cpu(rinfo->diri.in->snapid));
1187
1188                 /* do we have a lease on the whole dir? */
1189                 have_dir_cap =
1190                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1191                          CEPH_CAP_FILE_SHARED);
1192
1193                 /* do we have a dn lease? */
1194                 have_lease = have_dir_cap ||
1195                         le32_to_cpu(rinfo->dlease->duration_ms);
1196                 if (!have_lease)
1197                         dout("fill_trace  no dentry lease or dir cap\n");
1198
1199                 /* rename? */
1200                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1201                         struct inode *olddir = req->r_old_dentry_dir;
1202                         BUG_ON(!olddir);
1203
1204                         dout(" src %p '%pd' dst %p '%pd'\n",
1205                              req->r_old_dentry,
1206                              req->r_old_dentry,
1207                              dn, dn);
1208                         dout("fill_trace doing d_move %p -> %p\n",
1209                              req->r_old_dentry, dn);
1210
1211                         /* d_move screws up sibling dentries' offsets */
1212                         ceph_dir_clear_ordered(dir);
1213                         ceph_dir_clear_ordered(olddir);
1214
1215                         d_move(req->r_old_dentry, dn);
1216                         dout(" src %p '%pd' dst %p '%pd'\n",
1217                              req->r_old_dentry,
1218                              req->r_old_dentry,
1219                              dn, dn);
1220
1221                         /* ensure target dentry is invalidated, despite
1222                            rehashing bug in vfs_rename_dir */
1223                         ceph_invalidate_dentry_lease(dn);
1224
1225                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1226                              ceph_dentry(req->r_old_dentry)->offset);
1227
1228                         dn = req->r_old_dentry;  /* use old_dentry */
1229                 }
1230
1231                 /* null dentry? */
1232                 if (!rinfo->head->is_target) {
1233                         dout("fill_trace null dentry\n");
1234                         if (d_really_is_positive(dn)) {
1235                                 ceph_dir_clear_ordered(dir);
1236                                 dout("d_delete %p\n", dn);
1237                                 d_delete(dn);
1238                         } else {
1239                                 dout("d_instantiate %p NULL\n", dn);
1240                                 d_instantiate(dn, NULL);
1241                                 if (have_lease && d_unhashed(dn))
1242                                         d_rehash(dn);
1243                                 update_dentry_lease(dn, rinfo->dlease,
1244                                                     session,
1245                                                     req->r_request_started);
1246                         }
1247                         goto done;
1248                 }
1249
1250                 /* attach proper inode */
1251                 if (d_really_is_negative(dn)) {
1252                         ceph_dir_clear_ordered(dir);
1253                         ihold(in);
1254                         dn = splice_dentry(dn, in);
1255                         if (IS_ERR(dn)) {
1256                                 err = PTR_ERR(dn);
1257                                 goto done;
1258                         }
1259                         req->r_dentry = dn;  /* may have spliced */
1260                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1261                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1262                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1263                              ceph_vinop(in));
1264                         have_lease = false;
1265                 }
1266
1267                 if (have_lease)
1268                         update_dentry_lease(dn, rinfo->dlease, session,
1269                                             req->r_request_started);
1270                 dout(" final dn %p\n", dn);
1271         } else if (!req->r_aborted &&
1272                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1273                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1274                 struct dentry *dn = req->r_dentry;
1275                 struct inode *dir = req->r_locked_dir;
1276
1277                 /* fill out a snapdir LOOKUPSNAP dentry */
1278                 BUG_ON(!dn);
1279                 BUG_ON(!dir);
1280                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1281                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1282                 ceph_dir_clear_ordered(dir);
1283                 ihold(in);
1284                 dn = splice_dentry(dn, in);
1285                 if (IS_ERR(dn)) {
1286                         err = PTR_ERR(dn);
1287                         goto done;
1288                 }
1289                 req->r_dentry = dn;  /* may have spliced */
1290         }
1291 done:
1292         dout("fill_trace done err=%d\n", err);
1293         return err;
1294 }
1295
1296 /*
1297  * Prepopulate our cache with readdir results, leases, etc.
1298  */
1299 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1300                                            struct ceph_mds_session *session)
1301 {
1302         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1303         int i, err = 0;
1304
1305         for (i = 0; i < rinfo->dir_nr; i++) {
1306                 struct ceph_vino vino;
1307                 struct inode *in;
1308                 int rc;
1309
1310                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1311                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1312
1313                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1314                 if (IS_ERR(in)) {
1315                         err = PTR_ERR(in);
1316                         dout("new_inode badness got %d\n", err);
1317                         continue;
1318                 }
1319                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1320                                 req->r_request_started, -1,
1321                                 &req->r_caps_reservation);
1322                 if (rc < 0) {
1323                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1324                         err = rc;
1325                         continue;
1326                 }
1327         }
1328
1329         return err;
1330 }
1331
1332 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1333 {
1334         if (ctl->page) {
1335                 kunmap(ctl->page);
1336                 page_cache_release(ctl->page);
1337                 ctl->page = NULL;
1338         }
1339 }
1340
1341 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1342                               struct ceph_readdir_cache_control *ctl,
1343                               struct ceph_mds_request *req)
1344 {
1345         struct ceph_inode_info *ci = ceph_inode(dir);
1346         unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry*);
1347         unsigned idx = ctl->index % nsize;
1348         pgoff_t pgoff = ctl->index / nsize;
1349
1350         if (!ctl->page || pgoff != page_index(ctl->page)) {
1351                 ceph_readdir_cache_release(ctl);
1352                 ctl->page  = grab_cache_page(&dir->i_data, pgoff);
1353                 if (!ctl->page) {
1354                         ctl->index = -1;
1355                         return -ENOMEM;
1356                 }
1357                 /* reading/filling the cache are serialized by
1358                  * i_mutex, no need to use page lock */
1359                 unlock_page(ctl->page);
1360                 ctl->dentries = kmap(ctl->page);
1361         }
1362
1363         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1364             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1365                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1366                 ctl->dentries[idx] = dn;
1367                 ctl->index++;
1368         } else {
1369                 dout("disable readdir cache\n");
1370                 ctl->index = -1;
1371         }
1372         return 0;
1373 }
1374
1375 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1376                              struct ceph_mds_session *session)
1377 {
1378         struct dentry *parent = req->r_dentry;
1379         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1380         struct qstr dname;
1381         struct dentry *dn;
1382         struct inode *in;
1383         int err = 0, ret, i;
1384         struct inode *snapdir = NULL;
1385         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1386         struct ceph_dentry_info *di;
1387         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1388         struct ceph_readdir_cache_control cache_ctl = {};
1389
1390         if (req->r_aborted)
1391                 return readdir_prepopulate_inodes_only(req, session);
1392
1393         if (rinfo->dir_dir &&
1394             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1395                 dout("readdir_prepopulate got new frag %x -> %x\n",
1396                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1397                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1398                 if (ceph_frag_is_leftmost(frag))
1399                         req->r_readdir_offset = 2;
1400                 else
1401                         req->r_readdir_offset = 0;
1402         }
1403
1404         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1405                 snapdir = ceph_get_snapdir(d_inode(parent));
1406                 parent = d_find_alias(snapdir);
1407                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1408                      rinfo->dir_nr, parent);
1409         } else {
1410                 dout("readdir_prepopulate %d items under dn %p\n",
1411                      rinfo->dir_nr, parent);
1412                 if (rinfo->dir_dir)
1413                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1414         }
1415
1416         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1417                 /* note dir version at start of readdir so we can tell
1418                  * if any dentries get dropped */
1419                 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1420                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1421                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1422                 req->r_readdir_cache_idx = 0;
1423         }
1424
1425         cache_ctl.index = req->r_readdir_cache_idx;
1426
1427         /* FIXME: release caps/leases if error occurs */
1428         for (i = 0; i < rinfo->dir_nr; i++) {
1429                 struct ceph_vino vino;
1430
1431                 dname.name = rinfo->dir_dname[i];
1432                 dname.len = rinfo->dir_dname_len[i];
1433                 dname.hash = full_name_hash(dname.name, dname.len);
1434
1435                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1436                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1437
1438 retry_lookup:
1439                 dn = d_lookup(parent, &dname);
1440                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1441                      parent, dname.len, dname.name, dn);
1442
1443                 if (!dn) {
1444                         dn = d_alloc(parent, &dname);
1445                         dout("d_alloc %p '%.*s' = %p\n", parent,
1446                              dname.len, dname.name, dn);
1447                         if (dn == NULL) {
1448                                 dout("d_alloc badness\n");
1449                                 err = -ENOMEM;
1450                                 goto out;
1451                         }
1452                         ret = ceph_init_dentry(dn);
1453                         if (ret < 0) {
1454                                 dput(dn);
1455                                 err = ret;
1456                                 goto out;
1457                         }
1458                 } else if (d_really_is_positive(dn) &&
1459                            (ceph_ino(d_inode(dn)) != vino.ino ||
1460                             ceph_snap(d_inode(dn)) != vino.snap)) {
1461                         dout(" dn %p points to wrong inode %p\n",
1462                              dn, d_inode(dn));
1463                         d_delete(dn);
1464                         dput(dn);
1465                         goto retry_lookup;
1466                 }
1467
1468                 /* inode */
1469                 if (d_really_is_positive(dn)) {
1470                         in = d_inode(dn);
1471                 } else {
1472                         in = ceph_get_inode(parent->d_sb, vino);
1473                         if (IS_ERR(in)) {
1474                                 dout("new_inode badness\n");
1475                                 d_drop(dn);
1476                                 dput(dn);
1477                                 err = PTR_ERR(in);
1478                                 goto out;
1479                         }
1480                 }
1481
1482                 ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1483                                  req->r_request_started, -1,
1484                                  &req->r_caps_reservation);
1485                 if (ret < 0) {
1486                         pr_err("fill_inode badness on %p\n", in);
1487                         if (d_really_is_negative(dn))
1488                                 iput(in);
1489                         d_drop(dn);
1490                         err = ret;
1491                         goto next_item;
1492                 }
1493
1494                 if (d_really_is_negative(dn)) {
1495                         struct dentry *realdn = splice_dentry(dn, in);
1496                         if (IS_ERR(realdn)) {
1497                                 err = PTR_ERR(realdn);
1498                                 d_drop(dn);
1499                                 dn = NULL;
1500                                 goto next_item;
1501                         }
1502                         dn = realdn;
1503                 }
1504
1505                 di = dn->d_fsdata;
1506                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1507
1508                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1509                                     req->r_session,
1510                                     req->r_request_started);
1511
1512                 if (err == 0 && cache_ctl.index >= 0) {
1513                         ret = fill_readdir_cache(d_inode(parent), dn,
1514                                                  &cache_ctl, req);
1515                         if (ret < 0)
1516                                 err = ret;
1517                 }
1518 next_item:
1519                 if (dn)
1520                         dput(dn);
1521         }
1522 out:
1523         if (err == 0) {
1524                 req->r_did_prepopulate = true;
1525                 req->r_readdir_cache_idx = cache_ctl.index;
1526         }
1527         ceph_readdir_cache_release(&cache_ctl);
1528         if (snapdir) {
1529                 iput(snapdir);
1530                 dput(parent);
1531         }
1532         dout("readdir_prepopulate done\n");
1533         return err;
1534 }
1535
1536 int ceph_inode_set_size(struct inode *inode, loff_t size)
1537 {
1538         struct ceph_inode_info *ci = ceph_inode(inode);
1539         int ret = 0;
1540
1541         spin_lock(&ci->i_ceph_lock);
1542         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1543         i_size_write(inode, size);
1544         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1545
1546         /* tell the MDS if we are approaching max_size */
1547         if ((size << 1) >= ci->i_max_size &&
1548             (ci->i_reported_size << 1) < ci->i_max_size)
1549                 ret = 1;
1550
1551         spin_unlock(&ci->i_ceph_lock);
1552         return ret;
1553 }
1554
1555 /*
1556  * Write back inode data in a worker thread.  (This can't be done
1557  * in the message handler context.)
1558  */
1559 void ceph_queue_writeback(struct inode *inode)
1560 {
1561         ihold(inode);
1562         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1563                        &ceph_inode(inode)->i_wb_work)) {
1564                 dout("ceph_queue_writeback %p\n", inode);
1565         } else {
1566                 dout("ceph_queue_writeback %p failed\n", inode);
1567                 iput(inode);
1568         }
1569 }
1570
1571 static void ceph_writeback_work(struct work_struct *work)
1572 {
1573         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1574                                                   i_wb_work);
1575         struct inode *inode = &ci->vfs_inode;
1576
1577         dout("writeback %p\n", inode);
1578         filemap_fdatawrite(&inode->i_data);
1579         iput(inode);
1580 }
1581
1582 /*
1583  * queue an async invalidation
1584  */
1585 void ceph_queue_invalidate(struct inode *inode)
1586 {
1587         ihold(inode);
1588         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1589                        &ceph_inode(inode)->i_pg_inv_work)) {
1590                 dout("ceph_queue_invalidate %p\n", inode);
1591         } else {
1592                 dout("ceph_queue_invalidate %p failed\n", inode);
1593                 iput(inode);
1594         }
1595 }
1596
1597 /*
1598  * Invalidate inode pages in a worker thread.  (This can't be done
1599  * in the message handler context.)
1600  */
1601 static void ceph_invalidate_work(struct work_struct *work)
1602 {
1603         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1604                                                   i_pg_inv_work);
1605         struct inode *inode = &ci->vfs_inode;
1606         u32 orig_gen;
1607         int check = 0;
1608
1609         mutex_lock(&ci->i_truncate_mutex);
1610         spin_lock(&ci->i_ceph_lock);
1611         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1612              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1613         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1614                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1615                         check = 1;
1616                 spin_unlock(&ci->i_ceph_lock);
1617                 mutex_unlock(&ci->i_truncate_mutex);
1618                 goto out;
1619         }
1620         orig_gen = ci->i_rdcache_gen;
1621         spin_unlock(&ci->i_ceph_lock);
1622
1623         truncate_pagecache(inode, 0);
1624
1625         spin_lock(&ci->i_ceph_lock);
1626         if (orig_gen == ci->i_rdcache_gen &&
1627             orig_gen == ci->i_rdcache_revoking) {
1628                 dout("invalidate_pages %p gen %d successful\n", inode,
1629                      ci->i_rdcache_gen);
1630                 ci->i_rdcache_revoking--;
1631                 check = 1;
1632         } else {
1633                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1634                      inode, orig_gen, ci->i_rdcache_gen,
1635                      ci->i_rdcache_revoking);
1636                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1637                         check = 1;
1638         }
1639         spin_unlock(&ci->i_ceph_lock);
1640         mutex_unlock(&ci->i_truncate_mutex);
1641 out:
1642         if (check)
1643                 ceph_check_caps(ci, 0, NULL);
1644         iput(inode);
1645 }
1646
1647
1648 /*
1649  * called by trunc_wq;
1650  *
1651  * We also truncate in a separate thread as well.
1652  */
1653 static void ceph_vmtruncate_work(struct work_struct *work)
1654 {
1655         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1656                                                   i_vmtruncate_work);
1657         struct inode *inode = &ci->vfs_inode;
1658
1659         dout("vmtruncate_work %p\n", inode);
1660         __ceph_do_pending_vmtruncate(inode);
1661         iput(inode);
1662 }
1663
1664 /*
1665  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1666  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1667  */
1668 void ceph_queue_vmtruncate(struct inode *inode)
1669 {
1670         struct ceph_inode_info *ci = ceph_inode(inode);
1671
1672         ihold(inode);
1673
1674         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1675                        &ci->i_vmtruncate_work)) {
1676                 dout("ceph_queue_vmtruncate %p\n", inode);
1677         } else {
1678                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1679                      inode, ci->i_truncate_pending);
1680                 iput(inode);
1681         }
1682 }
1683
1684 /*
1685  * Make sure any pending truncation is applied before doing anything
1686  * that may depend on it.
1687  */
1688 void __ceph_do_pending_vmtruncate(struct inode *inode)
1689 {
1690         struct ceph_inode_info *ci = ceph_inode(inode);
1691         u64 to;
1692         int wrbuffer_refs, finish = 0;
1693
1694         mutex_lock(&ci->i_truncate_mutex);
1695 retry:
1696         spin_lock(&ci->i_ceph_lock);
1697         if (ci->i_truncate_pending == 0) {
1698                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1699                 spin_unlock(&ci->i_ceph_lock);
1700                 mutex_unlock(&ci->i_truncate_mutex);
1701                 return;
1702         }
1703
1704         /*
1705          * make sure any dirty snapped pages are flushed before we
1706          * possibly truncate them.. so write AND block!
1707          */
1708         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1709                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1710                      inode);
1711                 spin_unlock(&ci->i_ceph_lock);
1712                 filemap_write_and_wait_range(&inode->i_data, 0,
1713                                              inode->i_sb->s_maxbytes);
1714                 goto retry;
1715         }
1716
1717         /* there should be no reader or writer */
1718         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1719
1720         to = ci->i_truncate_size;
1721         wrbuffer_refs = ci->i_wrbuffer_ref;
1722         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1723              ci->i_truncate_pending, to);
1724         spin_unlock(&ci->i_ceph_lock);
1725
1726         truncate_pagecache(inode, to);
1727
1728         spin_lock(&ci->i_ceph_lock);
1729         if (to == ci->i_truncate_size) {
1730                 ci->i_truncate_pending = 0;
1731                 finish = 1;
1732         }
1733         spin_unlock(&ci->i_ceph_lock);
1734         if (!finish)
1735                 goto retry;
1736
1737         mutex_unlock(&ci->i_truncate_mutex);
1738
1739         if (wrbuffer_refs == 0)
1740                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1741
1742         wake_up_all(&ci->i_cap_wq);
1743 }
1744
1745 /*
1746  * symlinks
1747  */
1748 static const struct inode_operations ceph_symlink_iops = {
1749         .readlink = generic_readlink,
1750         .get_link = simple_get_link,
1751         .setattr = ceph_setattr,
1752         .getattr = ceph_getattr,
1753         .setxattr = ceph_setxattr,
1754         .getxattr = ceph_getxattr,
1755         .listxattr = ceph_listxattr,
1756         .removexattr = ceph_removexattr,
1757 };
1758
1759 /*
1760  * setattr
1761  */
1762 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1763 {
1764         struct inode *inode = d_inode(dentry);
1765         struct ceph_inode_info *ci = ceph_inode(inode);
1766         const unsigned int ia_valid = attr->ia_valid;
1767         struct ceph_mds_request *req;
1768         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1769         struct ceph_cap_flush *prealloc_cf;
1770         int issued;
1771         int release = 0, dirtied = 0;
1772         int mask = 0;
1773         int err = 0;
1774         int inode_dirty_flags = 0;
1775         bool lock_snap_rwsem = false;
1776
1777         if (ceph_snap(inode) != CEPH_NOSNAP)
1778                 return -EROFS;
1779
1780         err = inode_change_ok(inode, attr);
1781         if (err != 0)
1782                 return err;
1783
1784         prealloc_cf = ceph_alloc_cap_flush();
1785         if (!prealloc_cf)
1786                 return -ENOMEM;
1787
1788         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1789                                        USE_AUTH_MDS);
1790         if (IS_ERR(req)) {
1791                 ceph_free_cap_flush(prealloc_cf);
1792                 return PTR_ERR(req);
1793         }
1794
1795         spin_lock(&ci->i_ceph_lock);
1796         issued = __ceph_caps_issued(ci, NULL);
1797
1798         if (!ci->i_head_snapc &&
1799             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1800                 lock_snap_rwsem = true;
1801                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1802                         spin_unlock(&ci->i_ceph_lock);
1803                         down_read(&mdsc->snap_rwsem);
1804                         spin_lock(&ci->i_ceph_lock);
1805                         issued = __ceph_caps_issued(ci, NULL);
1806                 }
1807         }
1808
1809         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1810
1811         if (ia_valid & ATTR_UID) {
1812                 dout("setattr %p uid %d -> %d\n", inode,
1813                      from_kuid(&init_user_ns, inode->i_uid),
1814                      from_kuid(&init_user_ns, attr->ia_uid));
1815                 if (issued & CEPH_CAP_AUTH_EXCL) {
1816                         inode->i_uid = attr->ia_uid;
1817                         dirtied |= CEPH_CAP_AUTH_EXCL;
1818                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1819                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1820                         req->r_args.setattr.uid = cpu_to_le32(
1821                                 from_kuid(&init_user_ns, attr->ia_uid));
1822                         mask |= CEPH_SETATTR_UID;
1823                         release |= CEPH_CAP_AUTH_SHARED;
1824                 }
1825         }
1826         if (ia_valid & ATTR_GID) {
1827                 dout("setattr %p gid %d -> %d\n", inode,
1828                      from_kgid(&init_user_ns, inode->i_gid),
1829                      from_kgid(&init_user_ns, attr->ia_gid));
1830                 if (issued & CEPH_CAP_AUTH_EXCL) {
1831                         inode->i_gid = attr->ia_gid;
1832                         dirtied |= CEPH_CAP_AUTH_EXCL;
1833                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1834                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1835                         req->r_args.setattr.gid = cpu_to_le32(
1836                                 from_kgid(&init_user_ns, attr->ia_gid));
1837                         mask |= CEPH_SETATTR_GID;
1838                         release |= CEPH_CAP_AUTH_SHARED;
1839                 }
1840         }
1841         if (ia_valid & ATTR_MODE) {
1842                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1843                      attr->ia_mode);
1844                 if (issued & CEPH_CAP_AUTH_EXCL) {
1845                         inode->i_mode = attr->ia_mode;
1846                         dirtied |= CEPH_CAP_AUTH_EXCL;
1847                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1848                            attr->ia_mode != inode->i_mode) {
1849                         inode->i_mode = attr->ia_mode;
1850                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1851                         mask |= CEPH_SETATTR_MODE;
1852                         release |= CEPH_CAP_AUTH_SHARED;
1853                 }
1854         }
1855
1856         if (ia_valid & ATTR_ATIME) {
1857                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1858                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1859                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1860                 if (issued & CEPH_CAP_FILE_EXCL) {
1861                         ci->i_time_warp_seq++;
1862                         inode->i_atime = attr->ia_atime;
1863                         dirtied |= CEPH_CAP_FILE_EXCL;
1864                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1865                            timespec_compare(&inode->i_atime,
1866                                             &attr->ia_atime) < 0) {
1867                         inode->i_atime = attr->ia_atime;
1868                         dirtied |= CEPH_CAP_FILE_WR;
1869                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1870                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1871                         ceph_encode_timespec(&req->r_args.setattr.atime,
1872                                              &attr->ia_atime);
1873                         mask |= CEPH_SETATTR_ATIME;
1874                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1875                                 CEPH_CAP_FILE_WR;
1876                 }
1877         }
1878         if (ia_valid & ATTR_MTIME) {
1879                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1880                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1881                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1882                 if (issued & CEPH_CAP_FILE_EXCL) {
1883                         ci->i_time_warp_seq++;
1884                         inode->i_mtime = attr->ia_mtime;
1885                         dirtied |= CEPH_CAP_FILE_EXCL;
1886                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1887                            timespec_compare(&inode->i_mtime,
1888                                             &attr->ia_mtime) < 0) {
1889                         inode->i_mtime = attr->ia_mtime;
1890                         dirtied |= CEPH_CAP_FILE_WR;
1891                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1892                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1893                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1894                                              &attr->ia_mtime);
1895                         mask |= CEPH_SETATTR_MTIME;
1896                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1897                                 CEPH_CAP_FILE_WR;
1898                 }
1899         }
1900         if (ia_valid & ATTR_SIZE) {
1901                 dout("setattr %p size %lld -> %lld\n", inode,
1902                      inode->i_size, attr->ia_size);
1903                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1904                     attr->ia_size > inode->i_size) {
1905                         i_size_write(inode, attr->ia_size);
1906                         inode->i_blocks =
1907                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1908                         inode->i_ctime = attr->ia_ctime;
1909                         ci->i_reported_size = attr->ia_size;
1910                         dirtied |= CEPH_CAP_FILE_EXCL;
1911                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1912                            attr->ia_size != inode->i_size) {
1913                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1914                         req->r_args.setattr.old_size =
1915                                 cpu_to_le64(inode->i_size);
1916                         mask |= CEPH_SETATTR_SIZE;
1917                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1918                                 CEPH_CAP_FILE_WR;
1919                 }
1920         }
1921
1922         /* these do nothing */
1923         if (ia_valid & ATTR_CTIME) {
1924                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1925                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1926                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1927                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1928                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1929                      only ? "ctime only" : "ignored");
1930                 inode->i_ctime = attr->ia_ctime;
1931                 if (only) {
1932                         /*
1933                          * if kernel wants to dirty ctime but nothing else,
1934                          * we need to choose a cap to dirty under, or do
1935                          * a almost-no-op setattr
1936                          */
1937                         if (issued & CEPH_CAP_AUTH_EXCL)
1938                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1939                         else if (issued & CEPH_CAP_FILE_EXCL)
1940                                 dirtied |= CEPH_CAP_FILE_EXCL;
1941                         else if (issued & CEPH_CAP_XATTR_EXCL)
1942                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1943                         else
1944                                 mask |= CEPH_SETATTR_CTIME;
1945                 }
1946         }
1947         if (ia_valid & ATTR_FILE)
1948                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1949
1950         if (dirtied) {
1951                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
1952                                                            &prealloc_cf);
1953                 inode->i_ctime = CURRENT_TIME;
1954         }
1955
1956         release &= issued;
1957         spin_unlock(&ci->i_ceph_lock);
1958         if (lock_snap_rwsem)
1959                 up_read(&mdsc->snap_rwsem);
1960
1961         if (inode_dirty_flags)
1962                 __mark_inode_dirty(inode, inode_dirty_flags);
1963
1964         if (ia_valid & ATTR_MODE) {
1965                 err = posix_acl_chmod(inode, attr->ia_mode);
1966                 if (err)
1967                         goto out_put;
1968         }
1969
1970         if (mask) {
1971                 req->r_inode = inode;
1972                 ihold(inode);
1973                 req->r_inode_drop = release;
1974                 req->r_args.setattr.mask = cpu_to_le32(mask);
1975                 req->r_num_caps = 1;
1976                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1977         }
1978         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1979              ceph_cap_string(dirtied), mask);
1980
1981         ceph_mdsc_put_request(req);
1982         if (mask & CEPH_SETATTR_SIZE)
1983                 __ceph_do_pending_vmtruncate(inode);
1984         ceph_free_cap_flush(prealloc_cf);
1985         return err;
1986 out_put:
1987         ceph_mdsc_put_request(req);
1988         ceph_free_cap_flush(prealloc_cf);
1989         return err;
1990 }
1991
1992 /*
1993  * Verify that we have a lease on the given mask.  If not,
1994  * do a getattr against an mds.
1995  */
1996 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1997                       int mask, bool force)
1998 {
1999         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2000         struct ceph_mds_client *mdsc = fsc->mdsc;
2001         struct ceph_mds_request *req;
2002         int err;
2003
2004         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2005                 dout("do_getattr inode %p SNAPDIR\n", inode);
2006                 return 0;
2007         }
2008
2009         dout("do_getattr inode %p mask %s mode 0%o\n",
2010              inode, ceph_cap_string(mask), inode->i_mode);
2011         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2012                 return 0;
2013
2014         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2015         if (IS_ERR(req))
2016                 return PTR_ERR(req);
2017         req->r_inode = inode;
2018         ihold(inode);
2019         req->r_num_caps = 1;
2020         req->r_args.getattr.mask = cpu_to_le32(mask);
2021         req->r_locked_page = locked_page;
2022         err = ceph_mdsc_do_request(mdsc, NULL, req);
2023         if (locked_page && err == 0) {
2024                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2025                 if (inline_version == 0) {
2026                         /* the reply is supposed to contain inline data */
2027                         err = -EINVAL;
2028                 } else if (inline_version == CEPH_INLINE_NONE) {
2029                         err = -ENODATA;
2030                 } else {
2031                         err = req->r_reply_info.targeti.inline_len;
2032                 }
2033         }
2034         ceph_mdsc_put_request(req);
2035         dout("do_getattr result=%d\n", err);
2036         return err;
2037 }
2038
2039
2040 /*
2041  * Check inode permissions.  We verify we have a valid value for
2042  * the AUTH cap, then call the generic handler.
2043  */
2044 int ceph_permission(struct inode *inode, int mask)
2045 {
2046         int err;
2047
2048         if (mask & MAY_NOT_BLOCK)
2049                 return -ECHILD;
2050
2051         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2052
2053         if (!err)
2054                 err = generic_permission(inode, mask);
2055         return err;
2056 }
2057
2058 /*
2059  * Get all attributes.  Hopefully somedata we'll have a statlite()
2060  * and can limit the fields we require to be accurate.
2061  */
2062 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2063                  struct kstat *stat)
2064 {
2065         struct inode *inode = d_inode(dentry);
2066         struct ceph_inode_info *ci = ceph_inode(inode);
2067         int err;
2068
2069         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2070         if (!err) {
2071                 generic_fillattr(inode, stat);
2072                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2073                 if (ceph_snap(inode) != CEPH_NOSNAP)
2074                         stat->dev = ceph_snap(inode);
2075                 else
2076                         stat->dev = 0;
2077                 if (S_ISDIR(inode->i_mode)) {
2078                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2079                                                 RBYTES))
2080                                 stat->size = ci->i_rbytes;
2081                         else
2082                                 stat->size = ci->i_files + ci->i_subdirs;
2083                         stat->blocks = 0;
2084                         stat->blksize = 65536;
2085                 }
2086         }
2087         return err;
2088 }