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_snapdir_iops;
  86         inode->i_fop = &ceph_snapdir_fops;
  87         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  88         ci->i_rbytes = 0;
  89         return inode;
  90 }
  91
  92 const struct inode_operations ceph_file_iops = {
  93         .permission = ceph_permission,
  94         .setattr = ceph_setattr,
  95         .getattr = ceph_getattr,
  96         .setxattr = ceph_setxattr,
  97         .getxattr = ceph_getxattr,
  98         .listxattr = ceph_listxattr,
  99         .removexattr = ceph_removexattr,
 100         .get_acl = ceph_get_acl,
 101         .set_acl = ceph_set_acl,
 102 };
 103
 104
 105 /*
 106  * We use a 'frag tree' to keep track of the MDS's directory fragments
 107  * for a given inode (usually there is just a single fragment).  We
 108  * need to know when a child frag is delegated to a new MDS, or when
 109  * it is flagged as replicated, so we can direct our requests
 110  * accordingly.
 111  */
 112
 113 /*
 114  * find/create a frag in the tree
 115  */
 116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 117                                                     u32 f)
 118 {
 119         struct rb_node **p;
 120         struct rb_node *parent = NULL;
 121         struct ceph_inode_frag *frag;
 122         int c;
 123
 124         p = &ci->i_fragtree.rb_node;
 125         while (*p) {
 126                 parent = *p;
 127                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 128                 c = ceph_frag_compare(f, frag->frag);
 129                 if (c < 0)
 130                         p = &(*p)->rb_left;
 131                 else if (c > 0)
 132                         p = &(*p)->rb_right;
 133                 else
 134                         return frag;
 135         }
 136
 137         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 138         if (!frag) {
 139                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 140                        "frag %x\n", &ci->vfs_inode,
 141                        ceph_vinop(&ci->vfs_inode), f);
 142                 return ERR_PTR(-ENOMEM);
 143         }
 144         frag->frag = f;
 145         frag->split_by = 0;
 146         frag->mds = -1;
 147         frag->ndist = 0;
 148
 149         rb_link_node(&frag->node, parent, p);
 150         rb_insert_color(&frag->node, &ci->i_fragtree);
 151
 152         dout("get_or_create_frag added %llx.%llx frag %x\n",
 153              ceph_vinop(&ci->vfs_inode), f);
 154         return frag;
 155 }
 156
 157 /*
 158  * find a specific frag @f
 159  */
 160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 161 {
 162         struct rb_node *n = ci->i_fragtree.rb_node;
 163
 164         while (n) {
 165                 struct ceph_inode_frag *frag =
 166                         rb_entry(n, struct ceph_inode_frag, node);
 167                 int c = ceph_frag_compare(f, frag->frag);
 168                 if (c < 0)
 169                         n = n->rb_left;
 170                 else if (c > 0)
 171                         n = n->rb_right;
 172                 else
 173                         return frag;
 174         }
 175         return NULL;
 176 }
 177
 178 /*
 179  * Choose frag containing the given value @v.  If @pfrag is
 180  * specified, copy the frag delegation info to the caller if
 181  * it is present.
 182  */
 183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 184                               struct ceph_inode_frag *pfrag, int *found)
 185 {
 186         u32 t = ceph_frag_make(0, 0);
 187         struct ceph_inode_frag *frag;
 188         unsigned nway, i;
 189         u32 n;
 190
 191         if (found)
 192                 *found = 0;
 193
 194         while (1) {
 195                 WARN_ON(!ceph_frag_contains_value(t, v));
 196                 frag = __ceph_find_frag(ci, t);
 197                 if (!frag)
 198                         break; /* t is a leaf */
 199                 if (frag->split_by == 0) {
 200                         if (pfrag)
 201                                 memcpy(pfrag, frag, sizeof(*pfrag));
 202                         if (found)
 203                                 *found = 1;
 204                         break;
 205                 }
 206
 207                 /* choose child */
 208                 nway = 1 << frag->split_by;
 209                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 210                      frag->split_by, nway);
 211                 for (i = 0; i < nway; i++) {
 212                         n = ceph_frag_make_child(t, frag->split_by, i);
 213                         if (ceph_frag_contains_value(n, v)) {
 214                                 t = n;
 215                                 break;
 216                         }
 217                 }
 218                 BUG_ON(i == nway);
 219         }
 220         dout("choose_frag(%x) = %x\n", v, t);
 221
 222         return t;
 223 }
 224
 225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 226                      struct ceph_inode_frag *pfrag, int *found)
 227 {
 228         u32 ret;
 229         mutex_lock(&ci->i_fragtree_mutex);
 230         ret = __ceph_choose_frag(ci, v, pfrag, found);
 231         mutex_unlock(&ci->i_fragtree_mutex);
 232         return ret;
 233 }
 234
 235 /*
 236  * Process dirfrag (delegation) info from the mds.  Include leaf
 237  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 238  * branches/splits are included in i_fragtree)
 239  */
 240 static int ceph_fill_dirfrag(struct inode *inode,
 241                              struct ceph_mds_reply_dirfrag *dirinfo)
 242 {
 243         struct ceph_inode_info *ci = ceph_inode(inode);
 244         struct ceph_inode_frag *frag;
 245         u32 id = le32_to_cpu(dirinfo->frag);
 246         int mds = le32_to_cpu(dirinfo->auth);
 247         int ndist = le32_to_cpu(dirinfo->ndist);
 248         int diri_auth = -1;
 249         int i;
 250         int err = 0;
 251
 252         spin_lock(&ci->i_ceph_lock);
 253         if (ci->i_auth_cap)
 254                 diri_auth = ci->i_auth_cap->mds;
 255         spin_unlock(&ci->i_ceph_lock);
 256
 257         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_inline_version = 0;
 391         ci->i_time_warp_seq = 0;
 392         ci->i_ceph_flags = 0;
 393         ci->i_ordered_count = 0;
 394         atomic_set(&ci->i_release_count, 1);
 395         atomic_set(&ci->i_complete_count, 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_cap_flush_tree = RB_ROOT;
 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, struct page *locked_page,
 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         bool fill_inline = false;
 679
 680         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 681              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 682              ci->i_version);
 683
 684         /* prealloc new cap struct */
 685         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 686                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 687
 688         /*
 689          * prealloc xattr data, if it looks like we'll need it.  only
 690          * if len > 4 (meaning there are actually xattrs; the first 4
 691          * bytes are the xattr count).
 692          */
 693         if (iinfo->xattr_len > 4) {
 694                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 695                 if (!xattr_blob)
 696                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 697                                iinfo->xattr_len);
 698         }
 699
 700         spin_lock(&ci->i_ceph_lock);
 701
 702         /*
 703          * provided version will be odd if inode value is projected,
 704          * even if stable.  skip the update if we have newer stable
 705          * info (ours>=theirs, e.g. due to racing mds replies), unless
 706          * we are getting projected (unstable) info (in which case the
 707          * version is odd, and we want ours>theirs).
 708          *   us   them
 709          *   2    2     skip
 710          *   3    2     skip
 711          *   3    3     update
 712          */
 713         if (ci->i_version == 0 ||
 714             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 715              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 716                 new_version = true;
 717
 718         issued = __ceph_caps_issued(ci, &implemented);
 719         issued |= implemented | __ceph_caps_dirty(ci);
 720         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 721
 722         /* update inode */
 723         ci->i_version = le64_to_cpu(info->version);
 724         inode->i_version++;
 725         inode->i_rdev = le32_to_cpu(info->rdev);
 726         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 727
 728         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 729             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 730                 inode->i_mode = le32_to_cpu(info->mode);
 731                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 732                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 733                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 734                      from_kuid(&init_user_ns, inode->i_uid),
 735                      from_kgid(&init_user_ns, inode->i_gid));
 736         }
 737
 738         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 739             (issued & CEPH_CAP_LINK_EXCL) == 0)
 740                 set_nlink(inode, le32_to_cpu(info->nlink));
 741
 742         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 743                 /* be careful with mtime, atime, size */
 744                 ceph_decode_timespec(&atime, &info->atime);
 745                 ceph_decode_timespec(&mtime, &info->mtime);
 746                 ceph_decode_timespec(&ctime, &info->ctime);
 747                 ceph_fill_file_time(inode, issued,
 748                                 le32_to_cpu(info->time_warp_seq),
 749                                 &ctime, &mtime, &atime);
 750         }
 751
 752         if (new_version ||
 753             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 754                 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
 755                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 756                 ci->i_layout = info->layout;
 757
 758                 queue_trunc = ceph_fill_file_size(inode, issued,
 759                                         le32_to_cpu(info->truncate_seq),
 760                                         le64_to_cpu(info->truncate_size),
 761                                         le64_to_cpu(info->size));
 762                 /* only update max_size on auth cap */
 763                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 764                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 765                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 766                                         le64_to_cpu(info->max_size));
 767                         ci->i_max_size = le64_to_cpu(info->max_size);
 768                 }
 769         }
 770
 771         /* xattrs */
 772         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 773         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 774             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 775                 if (ci->i_xattrs.blob)
 776                         ceph_buffer_put(ci->i_xattrs.blob);
 777                 ci->i_xattrs.blob = xattr_blob;
 778                 if (xattr_blob)
 779                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 780                                iinfo->xattr_data, iinfo->xattr_len);
 781                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 782                 ceph_forget_all_cached_acls(inode);
 783                 xattr_blob = NULL;
 784         }
 785
 786         inode->i_mapping->a_ops = &ceph_aops;
 787
 788         switch (inode->i_mode & S_IFMT) {
 789         case S_IFIFO:
 790         case S_IFBLK:
 791         case S_IFCHR:
 792         case S_IFSOCK:
 793                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 794                 inode->i_op = &ceph_file_iops;
 795                 break;
 796         case S_IFREG:
 797                 inode->i_op = &ceph_file_iops;
 798                 inode->i_fop = &ceph_file_fops;
 799                 break;
 800         case S_IFLNK:
 801                 inode->i_op = &ceph_symlink_iops;
 802                 if (!ci->i_symlink) {
 803                         u32 symlen = iinfo->symlink_len;
 804                         char *sym;
 805
 806                         spin_unlock(&ci->i_ceph_lock);
 807
 808                         err = -EINVAL;
 809                         if (WARN_ON(symlen != inode->i_size))
 810                                 goto out;
 811
 812                         err = -ENOMEM;
 813                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 814                         if (!sym)
 815                                 goto out;
 816
 817                         spin_lock(&ci->i_ceph_lock);
 818                         if (!ci->i_symlink)
 819                                 ci->i_symlink = sym;
 820                         else
 821                                 kfree(sym); /* lost a race */
 822                 }
 823                 break;
 824         case S_IFDIR:
 825                 inode->i_op = &ceph_dir_iops;
 826                 inode->i_fop = &ceph_dir_fops;
 827
 828                 ci->i_dir_layout = iinfo->dir_layout;
 829
 830                 ci->i_files = le64_to_cpu(info->files);
 831                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 832                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 833                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 834                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 835                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 836                 break;
 837         default:
 838                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 839                        ceph_vinop(inode), inode->i_mode);
 840         }
 841
 842         /* were we issued a capability? */
 843         if (info->cap.caps) {
 844                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 845                         unsigned caps = le32_to_cpu(info->cap.caps);
 846                         ceph_add_cap(inode, session,
 847                                      le64_to_cpu(info->cap.cap_id),
 848                                      cap_fmode, caps,
 849                                      le32_to_cpu(info->cap.wanted),
 850                                      le32_to_cpu(info->cap.seq),
 851                                      le32_to_cpu(info->cap.mseq),
 852                                      le64_to_cpu(info->cap.realm),
 853                                      info->cap.flags, &new_cap);
 854
 855                         /* set dir completion flag? */
 856                         if (S_ISDIR(inode->i_mode) &&
 857                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 858                             (caps & CEPH_CAP_FILE_SHARED) &&
 859                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 860                             !__ceph_dir_is_complete(ci)) {
 861                                 dout(" marking %p complete (empty)\n", inode);
 862                                 __ceph_dir_set_complete(ci,
 863                                         atomic_read(&ci->i_release_count),
 864                                         ci->i_ordered_count);
 865                         }
 866
 867                         wake = true;
 868                 } else {
 869                         dout(" %p got snap_caps %s\n", inode,
 870                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 871                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 872                         if (cap_fmode >= 0)
 873                                 __ceph_get_fmode(ci, cap_fmode);
 874                 }
 875         } else if (cap_fmode >= 0) {
 876                 pr_warn("mds issued no caps on %llx.%llx\n",
 877                            ceph_vinop(inode));
 878                 __ceph_get_fmode(ci, cap_fmode);
 879         }
 880
 881         if (iinfo->inline_version > 0 &&
 882             iinfo->inline_version >= ci->i_inline_version) {
 883                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 884                 ci->i_inline_version = iinfo->inline_version;
 885                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 886                     (locked_page ||
 887                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 888                         fill_inline = true;
 889         }
 890
 891         spin_unlock(&ci->i_ceph_lock);
 892
 893         if (fill_inline)
 894                 ceph_fill_inline_data(inode, locked_page,
 895                                       iinfo->inline_data, iinfo->inline_len);
 896
 897         if (wake)
 898                 wake_up_all(&ci->i_cap_wq);
 899
 900         /* queue truncate if we saw i_size decrease */
 901         if (queue_trunc)
 902                 ceph_queue_vmtruncate(inode);
 903
 904         /* populate frag tree */
 905         if (S_ISDIR(inode->i_mode))
 906                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 907
 908         /* update delegation info? */
 909         if (dirinfo)
 910                 ceph_fill_dirfrag(inode, dirinfo);
 911
 912         err = 0;
 913 out:
 914         if (new_cap)
 915                 ceph_put_cap(mdsc, new_cap);
 916         if (xattr_blob)
 917                 ceph_buffer_put(xattr_blob);
 918         return err;
 919 }
 920
 921 /*
 922  * caller should hold session s_mutex.
 923  */
 924 static void update_dentry_lease(struct dentry *dentry,
 925                                 struct ceph_mds_reply_lease *lease,
 926                                 struct ceph_mds_session *session,
 927                                 unsigned long from_time)
 928 {
 929         struct ceph_dentry_info *di = ceph_dentry(dentry);
 930         long unsigned duration = le32_to_cpu(lease->duration_ms);
 931         long unsigned ttl = from_time + (duration * HZ) / 1000;
 932         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 933         struct inode *dir;
 934
 935         /* only track leases on regular dentries */
 936         if (dentry->d_op != &ceph_dentry_ops)
 937                 return;
 938
 939         spin_lock(&dentry->d_lock);
 940         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 941              dentry, duration, ttl);
 942
 943         /* make lease_rdcache_gen match directory */
 944         dir = d_inode(dentry->d_parent);
 945         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 946
 947         if (duration == 0)
 948                 goto out_unlock;
 949
 950         if (di->lease_gen == session->s_cap_gen &&
 951             time_before(ttl, dentry->d_time))
 952                 goto out_unlock;  /* we already have a newer lease. */
 953
 954         if (di->lease_session && di->lease_session != session)
 955                 goto out_unlock;
 956
 957         ceph_dentry_lru_touch(dentry);
 958
 959         if (!di->lease_session)
 960                 di->lease_session = ceph_get_mds_session(session);
 961         di->lease_gen = session->s_cap_gen;
 962         di->lease_seq = le32_to_cpu(lease->seq);
 963         di->lease_renew_after = half_ttl;
 964         di->lease_renew_from = 0;
 965         dentry->d_time = ttl;
 966 out_unlock:
 967         spin_unlock(&dentry->d_lock);
 968         return;
 969 }
 970
 971 /*
 972  * splice a dentry to an inode.
 973  * caller must hold directory i_mutex for this to be safe.
 974  *
 975  * we will only rehash the resulting dentry if @prehash is
 976  * true; @prehash will be set to false (for the benefit of
 977  * the caller) if we fail.
 978  */
 979 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 980                                     bool *prehash)
 981 {
 982         struct dentry *realdn;
 983
 984         BUG_ON(d_inode(dn));
 985
 986         /* dn must be unhashed */
 987         if (!d_unhashed(dn))
 988                 d_drop(dn);
 989         realdn = d_splice_alias(in, dn);
 990         if (IS_ERR(realdn)) {
 991                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 992                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 993                 if (prehash)
 994                         *prehash = false; /* don't rehash on error */
 995                 dn = realdn; /* note realdn contains the error */
 996                 goto out;
 997         } else if (realdn) {
 998                 dout("dn %p (%d) spliced with %p (%d) "
 999                      "inode %p ino %llx.%llx\n",
1000                      dn, d_count(dn),
1001                      realdn, d_count(realdn),
1002                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1003                 dput(dn);
1004                 dn = realdn;
1005         } else {
1006                 BUG_ON(!ceph_dentry(dn));
1007                 dout("dn %p attached to %p ino %llx.%llx\n",
1008                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1009         }
1010         if ((!prehash || *prehash) && d_unhashed(dn))
1011                 d_rehash(dn);
1012 out:
1013         return dn;
1014 }
1015
1016 /*
1017  * Incorporate results into the local cache.  This is either just
1018  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1019  * after a lookup).
1020  *
1021  * A reply may contain
1022  *         a directory inode along with a dentry.
1023  *  and/or a target inode
1024  *
1025  * Called with snap_rwsem (read).
1026  */
1027 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1028                     struct ceph_mds_session *session)
1029 {
1030         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1031         struct inode *in = NULL;
1032         struct ceph_vino vino;
1033         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1034         int err = 0;
1035
1036         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1037              rinfo->head->is_dentry, rinfo->head->is_target);
1038
1039 #if 0
1040         /*
1041          * Debugging hook:
1042          *
1043          * If we resend completed ops to a recovering mds, we get no
1044          * trace.  Since that is very rare, pretend this is the case
1045          * to ensure the 'no trace' handlers in the callers behave.
1046          *
1047          * Fill in inodes unconditionally to avoid breaking cap
1048          * invariants.
1049          */
1050         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1051                 pr_info("fill_trace faking empty trace on %lld %s\n",
1052                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1053                 if (rinfo->head->is_dentry) {
1054                         rinfo->head->is_dentry = 0;
1055                         err = fill_inode(req->r_locked_dir,
1056                                          &rinfo->diri, rinfo->dirfrag,
1057                                          session, req->r_request_started, -1);
1058                 }
1059                 if (rinfo->head->is_target) {
1060                         rinfo->head->is_target = 0;
1061                         ininfo = rinfo->targeti.in;
1062                         vino.ino = le64_to_cpu(ininfo->ino);
1063                         vino.snap = le64_to_cpu(ininfo->snapid);
1064                         in = ceph_get_inode(sb, vino);
1065                         err = fill_inode(in, &rinfo->targeti, NULL,
1066                                          session, req->r_request_started,
1067                                          req->r_fmode);
1068                         iput(in);
1069                 }
1070         }
1071 #endif
1072
1073         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1074                 dout("fill_trace reply is empty!\n");
1075                 if (rinfo->head->result == 0 && req->r_locked_dir)
1076                         ceph_invalidate_dir_request(req);
1077                 return 0;
1078         }
1079
1080         if (rinfo->head->is_dentry) {
1081                 struct inode *dir = req->r_locked_dir;
1082
1083                 if (dir) {
1084                         err = fill_inode(dir, NULL,
1085                                          &rinfo->diri, rinfo->dirfrag,
1086                                          session, req->r_request_started, -1,
1087                                          &req->r_caps_reservation);
1088                         if (err < 0)
1089                                 goto done;
1090                 } else {
1091                         WARN_ON_ONCE(1);
1092                 }
1093
1094                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1095                         struct qstr dname;
1096                         struct dentry *dn, *parent;
1097
1098                         BUG_ON(!rinfo->head->is_target);
1099                         BUG_ON(req->r_dentry);
1100
1101                         parent = d_find_any_alias(dir);
1102                         BUG_ON(!parent);
1103
1104                         dname.name = rinfo->dname;
1105                         dname.len = rinfo->dname_len;
1106                         dname.hash = full_name_hash(dname.name, dname.len);
1107                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1108                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1109 retry_lookup:
1110                         dn = d_lookup(parent, &dname);
1111                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1112                              parent, dname.len, dname.name, dn);
1113
1114                         if (!dn) {
1115                                 dn = d_alloc(parent, &dname);
1116                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1117                                      dname.len, dname.name, dn);
1118                                 if (dn == NULL) {
1119                                         dput(parent);
1120                                         err = -ENOMEM;
1121                                         goto done;
1122                                 }
1123                                 err = ceph_init_dentry(dn);
1124                                 if (err < 0) {
1125                                         dput(dn);
1126                                         dput(parent);
1127                                         goto done;
1128                                 }
1129                         } else if (d_really_is_positive(dn) &&
1130                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1131                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1132                                 dout(" dn %p points to wrong inode %p\n",
1133                                      dn, d_inode(dn));
1134                                 d_delete(dn);
1135                                 dput(dn);
1136                                 goto retry_lookup;
1137                         }
1138
1139                         req->r_dentry = dn;
1140                         dput(parent);
1141                 }
1142         }
1143
1144         if (rinfo->head->is_target) {
1145                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1146                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1147
1148                 in = ceph_get_inode(sb, vino);
1149                 if (IS_ERR(in)) {
1150                         err = PTR_ERR(in);
1151                         goto done;
1152                 }
1153                 req->r_target_inode = in;
1154
1155                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1156                                 session, req->r_request_started,
1157                                 (!req->r_aborted && rinfo->head->result == 0) ?
1158                                 req->r_fmode : -1,
1159                                 &req->r_caps_reservation);
1160                 if (err < 0) {
1161                         pr_err("fill_inode badness %p %llx.%llx\n",
1162                                 in, ceph_vinop(in));
1163                         goto done;
1164                 }
1165         }
1166
1167         /*
1168          * ignore null lease/binding on snapdir ENOENT, or else we
1169          * will have trouble splicing in the virtual snapdir later
1170          */
1171         if (rinfo->head->is_dentry && !req->r_aborted &&
1172             req->r_locked_dir &&
1173             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1174                                                fsc->mount_options->snapdir_name,
1175                                                req->r_dentry->d_name.len))) {
1176                 /*
1177                  * lookup link rename   : null -> possibly existing inode
1178                  * mknod symlink mkdir  : null -> new inode
1179                  * unlink               : linked -> null
1180                  */
1181                 struct inode *dir = req->r_locked_dir;
1182                 struct dentry *dn = req->r_dentry;
1183                 bool have_dir_cap, have_lease;
1184
1185                 BUG_ON(!dn);
1186                 BUG_ON(!dir);
1187                 BUG_ON(d_inode(dn->d_parent) != dir);
1188                 BUG_ON(ceph_ino(dir) !=
1189                        le64_to_cpu(rinfo->diri.in->ino));
1190                 BUG_ON(ceph_snap(dir) !=
1191                        le64_to_cpu(rinfo->diri.in->snapid));
1192
1193                 /* do we have a lease on the whole dir? */
1194                 have_dir_cap =
1195                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1196                          CEPH_CAP_FILE_SHARED);
1197
1198                 /* do we have a dn lease? */
1199                 have_lease = have_dir_cap ||
1200                         le32_to_cpu(rinfo->dlease->duration_ms);
1201                 if (!have_lease)
1202                         dout("fill_trace  no dentry lease or dir cap\n");
1203
1204                 /* rename? */
1205                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1206                         struct inode *olddir = req->r_old_dentry_dir;
1207                         BUG_ON(!olddir);
1208
1209                         dout(" src %p '%pd' dst %p '%pd'\n",
1210                              req->r_old_dentry,
1211                              req->r_old_dentry,
1212                              dn, dn);
1213                         dout("fill_trace doing d_move %p -> %p\n",
1214                              req->r_old_dentry, dn);
1215
1216                         d_move(req->r_old_dentry, dn);
1217                         dout(" src %p '%pd' dst %p '%pd'\n",
1218                              req->r_old_dentry,
1219                              req->r_old_dentry,
1220                              dn, dn);
1221
1222                         /* ensure target dentry is invalidated, despite
1223                            rehashing bug in vfs_rename_dir */
1224                         ceph_invalidate_dentry_lease(dn);
1225
1226                         /* d_move screws up sibling dentries' offsets */
1227                         ceph_dir_clear_ordered(dir);
1228                         ceph_dir_clear_ordered(olddir);
1229
1230                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1231                              ceph_dentry(req->r_old_dentry)->offset);
1232
1233                         dn = req->r_old_dentry;  /* use old_dentry */
1234                 }
1235
1236                 /* null dentry? */
1237                 if (!rinfo->head->is_target) {
1238                         dout("fill_trace null dentry\n");
1239                         if (d_really_is_positive(dn)) {
1240                                 ceph_dir_clear_ordered(dir);
1241                                 dout("d_delete %p\n", dn);
1242                                 d_delete(dn);
1243                         } else {
1244                                 dout("d_instantiate %p NULL\n", dn);
1245                                 d_instantiate(dn, NULL);
1246                                 if (have_lease && d_unhashed(dn))
1247                                         d_rehash(dn);
1248                                 update_dentry_lease(dn, rinfo->dlease,
1249                                                     session,
1250                                                     req->r_request_started);
1251                         }
1252                         goto done;
1253                 }
1254
1255                 /* attach proper inode */
1256                 if (d_really_is_negative(dn)) {
1257                         ceph_dir_clear_ordered(dir);
1258                         ihold(in);
1259                         dn = splice_dentry(dn, in, &have_lease);
1260                         if (IS_ERR(dn)) {
1261                                 err = PTR_ERR(dn);
1262                                 goto done;
1263                         }
1264                         req->r_dentry = dn;  /* may have spliced */
1265                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1266                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1267                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1268                              ceph_vinop(in));
1269                         have_lease = false;
1270                 }
1271
1272                 if (have_lease)
1273                         update_dentry_lease(dn, rinfo->dlease, session,
1274                                             req->r_request_started);
1275                 dout(" final dn %p\n", dn);
1276         } else if (!req->r_aborted &&
1277                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1278                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1279                 struct dentry *dn = req->r_dentry;
1280                 struct inode *dir = req->r_locked_dir;
1281
1282                 /* fill out a snapdir LOOKUPSNAP dentry */
1283                 BUG_ON(!dn);
1284                 BUG_ON(!dir);
1285                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1286                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1287                 ceph_dir_clear_ordered(dir);
1288                 ihold(in);
1289                 dn = splice_dentry(dn, in, NULL);
1290                 if (IS_ERR(dn)) {
1291                         err = PTR_ERR(dn);
1292                         goto done;
1293                 }
1294                 req->r_dentry = dn;  /* may have spliced */
1295         }
1296 done:
1297         dout("fill_trace done err=%d\n", err);
1298         return err;
1299 }
1300
1301 /*
1302  * Prepopulate our cache with readdir results, leases, etc.
1303  */
1304 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1305                                            struct ceph_mds_session *session)
1306 {
1307         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1308         int i, err = 0;
1309
1310         for (i = 0; i < rinfo->dir_nr; i++) {
1311                 struct ceph_vino vino;
1312                 struct inode *in;
1313                 int rc;
1314
1315                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1316                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1317
1318                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1319                 if (IS_ERR(in)) {
1320                         err = PTR_ERR(in);
1321                         dout("new_inode badness got %d\n", err);
1322                         continue;
1323                 }
1324                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1325                                 req->r_request_started, -1,
1326                                 &req->r_caps_reservation);
1327                 if (rc < 0) {
1328                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1329                         err = rc;
1330                         continue;
1331                 }
1332         }
1333
1334         return err;
1335 }
1336
1337 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1338                              struct ceph_mds_session *session)
1339 {
1340         struct dentry *parent = req->r_dentry;
1341         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1342         struct qstr dname;
1343         struct dentry *dn;
1344         struct inode *in;
1345         int err = 0, ret, i;
1346         struct inode *snapdir = NULL;
1347         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1348         struct ceph_dentry_info *di;
1349         u64 r_readdir_offset = req->r_readdir_offset;
1350         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1351
1352         if (rinfo->dir_dir &&
1353             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1354                 dout("readdir_prepopulate got new frag %x -> %x\n",
1355                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1356                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1357                 if (ceph_frag_is_leftmost(frag))
1358                         r_readdir_offset = 2;
1359                 else
1360                         r_readdir_offset = 0;
1361         }
1362
1363         if (req->r_aborted)
1364                 return readdir_prepopulate_inodes_only(req, session);
1365
1366         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1367                 snapdir = ceph_get_snapdir(d_inode(parent));
1368                 parent = d_find_alias(snapdir);
1369                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1370                      rinfo->dir_nr, parent);
1371         } else {
1372                 dout("readdir_prepopulate %d items under dn %p\n",
1373                      rinfo->dir_nr, parent);
1374                 if (rinfo->dir_dir)
1375                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1376         }
1377
1378         /* FIXME: release caps/leases if error occurs */
1379         for (i = 0; i < rinfo->dir_nr; i++) {
1380                 struct ceph_vino vino;
1381
1382                 dname.name = rinfo->dir_dname[i];
1383                 dname.len = rinfo->dir_dname_len[i];
1384                 dname.hash = full_name_hash(dname.name, dname.len);
1385
1386                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1387                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1388
1389 retry_lookup:
1390                 dn = d_lookup(parent, &dname);
1391                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1392                      parent, dname.len, dname.name, dn);
1393
1394                 if (!dn) {
1395                         dn = d_alloc(parent, &dname);
1396                         dout("d_alloc %p '%.*s' = %p\n", parent,
1397                              dname.len, dname.name, dn);
1398                         if (dn == NULL) {
1399                                 dout("d_alloc badness\n");
1400                                 err = -ENOMEM;
1401                                 goto out;
1402                         }
1403                         ret = ceph_init_dentry(dn);
1404                         if (ret < 0) {
1405                                 dput(dn);
1406                                 err = ret;
1407                                 goto out;
1408                         }
1409                 } else if (d_really_is_positive(dn) &&
1410                            (ceph_ino(d_inode(dn)) != vino.ino ||
1411                             ceph_snap(d_inode(dn)) != vino.snap)) {
1412                         dout(" dn %p points to wrong inode %p\n",
1413                              dn, d_inode(dn));
1414                         d_delete(dn);
1415                         dput(dn);
1416                         goto retry_lookup;
1417                 } else {
1418                         /* reorder parent's d_subdirs */
1419                         spin_lock(&parent->d_lock);
1420                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1421                         list_move(&dn->d_child, &parent->d_subdirs);
1422                         spin_unlock(&dn->d_lock);
1423                         spin_unlock(&parent->d_lock);
1424                 }
1425
1426                 /* inode */
1427                 if (d_really_is_positive(dn)) {
1428                         in = d_inode(dn);
1429                 } else {
1430                         in = ceph_get_inode(parent->d_sb, vino);
1431                         if (IS_ERR(in)) {
1432                                 dout("new_inode badness\n");
1433                                 d_drop(dn);
1434                                 dput(dn);
1435                                 err = PTR_ERR(in);
1436                                 goto out;
1437                         }
1438                 }
1439
1440                 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1441                                req->r_request_started, -1,
1442                                &req->r_caps_reservation) < 0) {
1443                         pr_err("fill_inode badness on %p\n", in);
1444                         if (d_really_is_negative(dn))
1445                                 iput(in);
1446                         d_drop(dn);
1447                         goto next_item;
1448                 }
1449
1450                 if (d_really_is_negative(dn)) {
1451                         struct dentry *realdn = splice_dentry(dn, in, NULL);
1452                         if (IS_ERR(realdn)) {
1453                                 err = PTR_ERR(realdn);
1454                                 d_drop(dn);
1455                                 dn = NULL;
1456                                 goto next_item;
1457                         }
1458                         dn = realdn;
1459                 }
1460
1461                 di = dn->d_fsdata;
1462                 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1463
1464                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1465                                     req->r_session,
1466                                     req->r_request_started);
1467 next_item:
1468                 if (dn)
1469                         dput(dn);
1470         }
1471         if (err == 0)
1472                 req->r_did_prepopulate = true;
1473
1474 out:
1475         if (snapdir) {
1476                 iput(snapdir);
1477                 dput(parent);
1478         }
1479         dout("readdir_prepopulate done\n");
1480         return err;
1481 }
1482
1483 int ceph_inode_set_size(struct inode *inode, loff_t size)
1484 {
1485         struct ceph_inode_info *ci = ceph_inode(inode);
1486         int ret = 0;
1487
1488         spin_lock(&ci->i_ceph_lock);
1489         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1490         inode->i_size = size;
1491         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1492
1493         /* tell the MDS if we are approaching max_size */
1494         if ((size << 1) >= ci->i_max_size &&
1495             (ci->i_reported_size << 1) < ci->i_max_size)
1496                 ret = 1;
1497
1498         spin_unlock(&ci->i_ceph_lock);
1499         return ret;
1500 }
1501
1502 /*
1503  * Write back inode data in a worker thread.  (This can't be done
1504  * in the message handler context.)
1505  */
1506 void ceph_queue_writeback(struct inode *inode)
1507 {
1508         ihold(inode);
1509         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1510                        &ceph_inode(inode)->i_wb_work)) {
1511                 dout("ceph_queue_writeback %p\n", inode);
1512         } else {
1513                 dout("ceph_queue_writeback %p failed\n", inode);
1514                 iput(inode);
1515         }
1516 }
1517
1518 static void ceph_writeback_work(struct work_struct *work)
1519 {
1520         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1521                                                   i_wb_work);
1522         struct inode *inode = &ci->vfs_inode;
1523
1524         dout("writeback %p\n", inode);
1525         filemap_fdatawrite(&inode->i_data);
1526         iput(inode);
1527 }
1528
1529 /*
1530  * queue an async invalidation
1531  */
1532 void ceph_queue_invalidate(struct inode *inode)
1533 {
1534         ihold(inode);
1535         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1536                        &ceph_inode(inode)->i_pg_inv_work)) {
1537                 dout("ceph_queue_invalidate %p\n", inode);
1538         } else {
1539                 dout("ceph_queue_invalidate %p failed\n", inode);
1540                 iput(inode);
1541         }
1542 }
1543
1544 /*
1545  * Invalidate inode pages in a worker thread.  (This can't be done
1546  * in the message handler context.)
1547  */
1548 static void ceph_invalidate_work(struct work_struct *work)
1549 {
1550         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1551                                                   i_pg_inv_work);
1552         struct inode *inode = &ci->vfs_inode;
1553         u32 orig_gen;
1554         int check = 0;
1555
1556         mutex_lock(&ci->i_truncate_mutex);
1557         spin_lock(&ci->i_ceph_lock);
1558         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1559              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1560         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1561                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1562                         check = 1;
1563                 spin_unlock(&ci->i_ceph_lock);
1564                 mutex_unlock(&ci->i_truncate_mutex);
1565                 goto out;
1566         }
1567         orig_gen = ci->i_rdcache_gen;
1568         spin_unlock(&ci->i_ceph_lock);
1569
1570         truncate_pagecache(inode, 0);
1571
1572         spin_lock(&ci->i_ceph_lock);
1573         if (orig_gen == ci->i_rdcache_gen &&
1574             orig_gen == ci->i_rdcache_revoking) {
1575                 dout("invalidate_pages %p gen %d successful\n", inode,
1576                      ci->i_rdcache_gen);
1577                 ci->i_rdcache_revoking--;
1578                 check = 1;
1579         } else {
1580                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1581                      inode, orig_gen, ci->i_rdcache_gen,
1582                      ci->i_rdcache_revoking);
1583                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1584                         check = 1;
1585         }
1586         spin_unlock(&ci->i_ceph_lock);
1587         mutex_unlock(&ci->i_truncate_mutex);
1588 out:
1589         if (check)
1590                 ceph_check_caps(ci, 0, NULL);
1591         iput(inode);
1592 }
1593
1594
1595 /*
1596  * called by trunc_wq;
1597  *
1598  * We also truncate in a separate thread as well.
1599  */
1600 static void ceph_vmtruncate_work(struct work_struct *work)
1601 {
1602         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1603                                                   i_vmtruncate_work);
1604         struct inode *inode = &ci->vfs_inode;
1605
1606         dout("vmtruncate_work %p\n", inode);
1607         __ceph_do_pending_vmtruncate(inode);
1608         iput(inode);
1609 }
1610
1611 /*
1612  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1613  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1614  */
1615 void ceph_queue_vmtruncate(struct inode *inode)
1616 {
1617         struct ceph_inode_info *ci = ceph_inode(inode);
1618
1619         ihold(inode);
1620
1621         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1622                        &ci->i_vmtruncate_work)) {
1623                 dout("ceph_queue_vmtruncate %p\n", inode);
1624         } else {
1625                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1626                      inode, ci->i_truncate_pending);
1627                 iput(inode);
1628         }
1629 }
1630
1631 /*
1632  * Make sure any pending truncation is applied before doing anything
1633  * that may depend on it.
1634  */
1635 void __ceph_do_pending_vmtruncate(struct inode *inode)
1636 {
1637         struct ceph_inode_info *ci = ceph_inode(inode);
1638         u64 to;
1639         int wrbuffer_refs, finish = 0;
1640
1641         mutex_lock(&ci->i_truncate_mutex);
1642 retry:
1643         spin_lock(&ci->i_ceph_lock);
1644         if (ci->i_truncate_pending == 0) {
1645                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1646                 spin_unlock(&ci->i_ceph_lock);
1647                 mutex_unlock(&ci->i_truncate_mutex);
1648                 return;
1649         }
1650
1651         /*
1652          * make sure any dirty snapped pages are flushed before we
1653          * possibly truncate them.. so write AND block!
1654          */
1655         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1656                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1657                      inode);
1658                 spin_unlock(&ci->i_ceph_lock);
1659                 filemap_write_and_wait_range(&inode->i_data, 0,
1660                                              inode->i_sb->s_maxbytes);
1661                 goto retry;
1662         }
1663
1664         /* there should be no reader or writer */
1665         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1666
1667         to = ci->i_truncate_size;
1668         wrbuffer_refs = ci->i_wrbuffer_ref;
1669         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1670              ci->i_truncate_pending, to);
1671         spin_unlock(&ci->i_ceph_lock);
1672
1673         truncate_pagecache(inode, to);
1674
1675         spin_lock(&ci->i_ceph_lock);
1676         if (to == ci->i_truncate_size) {
1677                 ci->i_truncate_pending = 0;
1678                 finish = 1;
1679         }
1680         spin_unlock(&ci->i_ceph_lock);
1681         if (!finish)
1682                 goto retry;
1683
1684         mutex_unlock(&ci->i_truncate_mutex);
1685
1686         if (wrbuffer_refs == 0)
1687                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1688
1689         wake_up_all(&ci->i_cap_wq);
1690 }
1691
1692 /*
1693  * symlinks
1694  */
1695 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1696 {
1697         struct ceph_inode_info *ci = ceph_inode(d_inode(dentry));
1698         nd_set_link(nd, ci->i_symlink);
1699         return NULL;
1700 }
1701
1702 static const struct inode_operations ceph_symlink_iops = {
1703         .readlink = generic_readlink,
1704         .follow_link = ceph_sym_follow_link,
1705         .setattr = ceph_setattr,
1706         .getattr = ceph_getattr,
1707         .setxattr = ceph_setxattr,
1708         .getxattr = ceph_getxattr,
1709         .listxattr = ceph_listxattr,
1710         .removexattr = ceph_removexattr,
1711 };
1712
1713 /*
1714  * setattr
1715  */
1716 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1717 {
1718         struct inode *inode = d_inode(dentry);
1719         struct ceph_inode_info *ci = ceph_inode(inode);
1720         const unsigned int ia_valid = attr->ia_valid;
1721         struct ceph_mds_request *req;
1722         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1723         int issued;
1724         int release = 0, dirtied = 0;
1725         int mask = 0;
1726         int err = 0;
1727         int inode_dirty_flags = 0;
1728         bool lock_snap_rwsem = false;
1729
1730         if (ceph_snap(inode) != CEPH_NOSNAP)
1731                 return -EROFS;
1732
1733         err = inode_change_ok(inode, attr);
1734         if (err != 0)
1735                 return err;
1736
1737         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1738                                        USE_AUTH_MDS);
1739         if (IS_ERR(req))
1740                 return PTR_ERR(req);
1741
1742         spin_lock(&ci->i_ceph_lock);
1743         issued = __ceph_caps_issued(ci, NULL);
1744
1745         if (!ci->i_head_snapc &&
1746             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1747                 lock_snap_rwsem = true;
1748                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1749                         spin_unlock(&ci->i_ceph_lock);
1750                         down_read(&mdsc->snap_rwsem);
1751                         spin_lock(&ci->i_ceph_lock);
1752                         issued = __ceph_caps_issued(ci, NULL);
1753                 }
1754         }
1755
1756         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1757
1758         if (ia_valid & ATTR_UID) {
1759                 dout("setattr %p uid %d -> %d\n", inode,
1760                      from_kuid(&init_user_ns, inode->i_uid),
1761                      from_kuid(&init_user_ns, attr->ia_uid));
1762                 if (issued & CEPH_CAP_AUTH_EXCL) {
1763                         inode->i_uid = attr->ia_uid;
1764                         dirtied |= CEPH_CAP_AUTH_EXCL;
1765                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1766                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1767                         req->r_args.setattr.uid = cpu_to_le32(
1768                                 from_kuid(&init_user_ns, attr->ia_uid));
1769                         mask |= CEPH_SETATTR_UID;
1770                         release |= CEPH_CAP_AUTH_SHARED;
1771                 }
1772         }
1773         if (ia_valid & ATTR_GID) {
1774                 dout("setattr %p gid %d -> %d\n", inode,
1775                      from_kgid(&init_user_ns, inode->i_gid),
1776                      from_kgid(&init_user_ns, attr->ia_gid));
1777                 if (issued & CEPH_CAP_AUTH_EXCL) {
1778                         inode->i_gid = attr->ia_gid;
1779                         dirtied |= CEPH_CAP_AUTH_EXCL;
1780                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1781                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1782                         req->r_args.setattr.gid = cpu_to_le32(
1783                                 from_kgid(&init_user_ns, attr->ia_gid));
1784                         mask |= CEPH_SETATTR_GID;
1785                         release |= CEPH_CAP_AUTH_SHARED;
1786                 }
1787         }
1788         if (ia_valid & ATTR_MODE) {
1789                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1790                      attr->ia_mode);
1791                 if (issued & CEPH_CAP_AUTH_EXCL) {
1792                         inode->i_mode = attr->ia_mode;
1793                         dirtied |= CEPH_CAP_AUTH_EXCL;
1794                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1795                            attr->ia_mode != inode->i_mode) {
1796                         inode->i_mode = attr->ia_mode;
1797                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1798                         mask |= CEPH_SETATTR_MODE;
1799                         release |= CEPH_CAP_AUTH_SHARED;
1800                 }
1801         }
1802
1803         if (ia_valid & ATTR_ATIME) {
1804                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1805                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1806                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1807                 if (issued & CEPH_CAP_FILE_EXCL) {
1808                         ci->i_time_warp_seq++;
1809                         inode->i_atime = attr->ia_atime;
1810                         dirtied |= CEPH_CAP_FILE_EXCL;
1811                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1812                            timespec_compare(&inode->i_atime,
1813                                             &attr->ia_atime) < 0) {
1814                         inode->i_atime = attr->ia_atime;
1815                         dirtied |= CEPH_CAP_FILE_WR;
1816                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1817                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1818                         ceph_encode_timespec(&req->r_args.setattr.atime,
1819                                              &attr->ia_atime);
1820                         mask |= CEPH_SETATTR_ATIME;
1821                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1822                                 CEPH_CAP_FILE_WR;
1823                 }
1824         }
1825         if (ia_valid & ATTR_MTIME) {
1826                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1827                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1828                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1829                 if (issued & CEPH_CAP_FILE_EXCL) {
1830                         ci->i_time_warp_seq++;
1831                         inode->i_mtime = attr->ia_mtime;
1832                         dirtied |= CEPH_CAP_FILE_EXCL;
1833                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1834                            timespec_compare(&inode->i_mtime,
1835                                             &attr->ia_mtime) < 0) {
1836                         inode->i_mtime = attr->ia_mtime;
1837                         dirtied |= CEPH_CAP_FILE_WR;
1838                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1839                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1840                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1841                                              &attr->ia_mtime);
1842                         mask |= CEPH_SETATTR_MTIME;
1843                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1844                                 CEPH_CAP_FILE_WR;
1845                 }
1846         }
1847         if (ia_valid & ATTR_SIZE) {
1848                 dout("setattr %p size %lld -> %lld\n", inode,
1849                      inode->i_size, attr->ia_size);
1850                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1851                     attr->ia_size > inode->i_size) {
1852                         inode->i_size = attr->ia_size;
1853                         inode->i_blocks =
1854                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1855                         inode->i_ctime = attr->ia_ctime;
1856                         ci->i_reported_size = attr->ia_size;
1857                         dirtied |= CEPH_CAP_FILE_EXCL;
1858                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1859                            attr->ia_size != inode->i_size) {
1860                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1861                         req->r_args.setattr.old_size =
1862                                 cpu_to_le64(inode->i_size);
1863                         mask |= CEPH_SETATTR_SIZE;
1864                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1865                                 CEPH_CAP_FILE_WR;
1866                 }
1867         }
1868
1869         /* these do nothing */
1870         if (ia_valid & ATTR_CTIME) {
1871                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1872                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1873                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1874                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1875                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1876                      only ? "ctime only" : "ignored");
1877                 inode->i_ctime = attr->ia_ctime;
1878                 if (only) {
1879                         /*
1880                          * if kernel wants to dirty ctime but nothing else,
1881                          * we need to choose a cap to dirty under, or do
1882                          * a almost-no-op setattr
1883                          */
1884                         if (issued & CEPH_CAP_AUTH_EXCL)
1885                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1886                         else if (issued & CEPH_CAP_FILE_EXCL)
1887                                 dirtied |= CEPH_CAP_FILE_EXCL;
1888                         else if (issued & CEPH_CAP_XATTR_EXCL)
1889                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1890                         else
1891                                 mask |= CEPH_SETATTR_CTIME;
1892                 }
1893         }
1894         if (ia_valid & ATTR_FILE)
1895                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1896
1897         if (dirtied) {
1898                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1899                 inode->i_ctime = CURRENT_TIME;
1900         }
1901
1902         release &= issued;
1903         spin_unlock(&ci->i_ceph_lock);
1904         if (lock_snap_rwsem)
1905                 up_read(&mdsc->snap_rwsem);
1906
1907         if (inode_dirty_flags)
1908                 __mark_inode_dirty(inode, inode_dirty_flags);
1909
1910         if (ia_valid & ATTR_MODE) {
1911                 err = posix_acl_chmod(inode, attr->ia_mode);
1912                 if (err)
1913                         goto out_put;
1914         }
1915
1916         if (mask) {
1917                 req->r_inode = inode;
1918                 ihold(inode);
1919                 req->r_inode_drop = release;
1920                 req->r_args.setattr.mask = cpu_to_le32(mask);
1921                 req->r_num_caps = 1;
1922                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1923         }
1924         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1925              ceph_cap_string(dirtied), mask);
1926
1927         ceph_mdsc_put_request(req);
1928         if (mask & CEPH_SETATTR_SIZE)
1929                 __ceph_do_pending_vmtruncate(inode);
1930         return err;
1931 out_put:
1932         ceph_mdsc_put_request(req);
1933         return err;
1934 }
1935
1936 /*
1937  * Verify that we have a lease on the given mask.  If not,
1938  * do a getattr against an mds.
1939  */
1940 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1941                       int mask, bool force)
1942 {
1943         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1944         struct ceph_mds_client *mdsc = fsc->mdsc;
1945         struct ceph_mds_request *req;
1946         int err;
1947
1948         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1949                 dout("do_getattr inode %p SNAPDIR\n", inode);
1950                 return 0;
1951         }
1952
1953         dout("do_getattr inode %p mask %s mode 0%o\n",
1954              inode, ceph_cap_string(mask), inode->i_mode);
1955         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1956                 return 0;
1957
1958         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1959         if (IS_ERR(req))
1960                 return PTR_ERR(req);
1961         req->r_inode = inode;
1962         ihold(inode);
1963         req->r_num_caps = 1;
1964         req->r_args.getattr.mask = cpu_to_le32(mask);
1965         req->r_locked_page = locked_page;
1966         err = ceph_mdsc_do_request(mdsc, NULL, req);
1967         if (locked_page && err == 0) {
1968                 u64 inline_version = req->r_reply_info.targeti.inline_version;
1969                 if (inline_version == 0) {
1970                         /* the reply is supposed to contain inline data */
1971                         err = -EINVAL;
1972                 } else if (inline_version == CEPH_INLINE_NONE) {
1973                         err = -ENODATA;
1974                 } else {
1975                         err = req->r_reply_info.targeti.inline_len;
1976                 }
1977         }
1978         ceph_mdsc_put_request(req);
1979         dout("do_getattr result=%d\n", err);
1980         return err;
1981 }
1982
1983
1984 /*
1985  * Check inode permissions.  We verify we have a valid value for
1986  * the AUTH cap, then call the generic handler.
1987  */
1988 int ceph_permission(struct inode *inode, int mask)
1989 {
1990         int err;
1991
1992         if (mask & MAY_NOT_BLOCK)
1993                 return -ECHILD;
1994
1995         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1996
1997         if (!err)
1998                 err = generic_permission(inode, mask);
1999         return err;
2000 }
2001
2002 /*
2003  * Get all attributes.  Hopefully somedata we'll have a statlite()
2004  * and can limit the fields we require to be accurate.
2005  */
2006 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2007                  struct kstat *stat)
2008 {
2009         struct inode *inode = d_inode(dentry);
2010         struct ceph_inode_info *ci = ceph_inode(inode);
2011         int err;
2012
2013         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2014         if (!err) {
2015                 generic_fillattr(inode, stat);
2016                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2017                 if (ceph_snap(inode) != CEPH_NOSNAP)
2018                         stat->dev = ceph_snap(inode);
2019                 else
2020                         stat->dev = 0;
2021                 if (S_ISDIR(inode->i_mode)) {
2022                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2023                                                 RBYTES))
2024                                 stat->size = ci->i_rbytes;
2025                         else
2026                                 stat->size = ci->i_files + ci->i_subdirs;
2027                         stat->blocks = 0;
2028                         stat->blksize = 65536;
2029                 }
2030         }
2031         return err;
2032 }