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