fs/ceph/inode.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/slab.h>
   6 #include <linux/string.h>
   7 #include <linux/uaccess.h>
   8 #include <linux/kernel.h>
   9 #include <linux/writeback.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/xattr.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 = generic_setxattr,
  97         .getxattr = generic_getxattr,
  98         .listxattr = ceph_listxattr,
  99         .removexattr = generic_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         atomic64_set(&ci->i_ordered_count, 1);
 394         atomic64_set(&ci->i_release_count, 1);
 395         atomic64_set(&ci->i_complete_seq[0], 0);
 396         atomic64_set(&ci->i_complete_seq[1], 0);
 397         ci->i_symlink = NULL;
 398
 399         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 400         ci->i_pool_ns_len = 0;
 401
 402         ci->i_fragtree = RB_ROOT;
 403         mutex_init(&ci->i_fragtree_mutex);
 404
 405         ci->i_xattrs.blob = NULL;
 406         ci->i_xattrs.prealloc_blob = NULL;
 407         ci->i_xattrs.dirty = false;
 408         ci->i_xattrs.index = RB_ROOT;
 409         ci->i_xattrs.count = 0;
 410         ci->i_xattrs.names_size = 0;
 411         ci->i_xattrs.vals_size = 0;
 412         ci->i_xattrs.version = 0;
 413         ci->i_xattrs.index_version = 0;
 414
 415         ci->i_caps = RB_ROOT;
 416         ci->i_auth_cap = NULL;
 417         ci->i_dirty_caps = 0;
 418         ci->i_flushing_caps = 0;
 419         INIT_LIST_HEAD(&ci->i_dirty_item);
 420         INIT_LIST_HEAD(&ci->i_flushing_item);
 421         ci->i_prealloc_cap_flush = NULL;
 422         ci->i_cap_flush_tree = RB_ROOT;
 423         init_waitqueue_head(&ci->i_cap_wq);
 424         ci->i_hold_caps_min = 0;
 425         ci->i_hold_caps_max = 0;
 426         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 427         INIT_LIST_HEAD(&ci->i_cap_snaps);
 428         ci->i_head_snapc = NULL;
 429         ci->i_snap_caps = 0;
 430
 431         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 432                 ci->i_nr_by_mode[i] = 0;
 433
 434         mutex_init(&ci->i_truncate_mutex);
 435         ci->i_truncate_seq = 0;
 436         ci->i_truncate_size = 0;
 437         ci->i_truncate_pending = 0;
 438
 439         ci->i_max_size = 0;
 440         ci->i_reported_size = 0;
 441         ci->i_wanted_max_size = 0;
 442         ci->i_requested_max_size = 0;
 443
 444         ci->i_pin_ref = 0;
 445         ci->i_rd_ref = 0;
 446         ci->i_rdcache_ref = 0;
 447         ci->i_wr_ref = 0;
 448         ci->i_wb_ref = 0;
 449         ci->i_wrbuffer_ref = 0;
 450         ci->i_wrbuffer_ref_head = 0;
 451         ci->i_shared_gen = 0;
 452         ci->i_rdcache_gen = 0;
 453         ci->i_rdcache_revoking = 0;
 454
 455         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 456         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 457         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 458         spin_lock_init(&ci->i_unsafe_lock);
 459
 460         ci->i_snap_realm = NULL;
 461         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 462         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 463
 464         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 465         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 466
 467         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 468
 469         ceph_fscache_inode_init(ci);
 470
 471         return &ci->vfs_inode;
 472 }
 473
 474 static void ceph_i_callback(struct rcu_head *head)
 475 {
 476         struct inode *inode = container_of(head, struct inode, i_rcu);
 477         struct ceph_inode_info *ci = ceph_inode(inode);
 478
 479         kmem_cache_free(ceph_inode_cachep, ci);
 480 }
 481
 482 void ceph_destroy_inode(struct inode *inode)
 483 {
 484         struct ceph_inode_info *ci = ceph_inode(inode);
 485         struct ceph_inode_frag *frag;
 486         struct rb_node *n;
 487
 488         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 489
 490         ceph_fscache_unregister_inode_cookie(ci);
 491
 492         ceph_queue_caps_release(inode);
 493
 494         /*
 495          * we may still have a snap_realm reference if there are stray
 496          * caps in i_snap_caps.
 497          */
 498         if (ci->i_snap_realm) {
 499                 struct ceph_mds_client *mdsc =
 500                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 501                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 502
 503                 dout(" dropping residual ref to snap realm %p\n", realm);
 504                 spin_lock(&realm->inodes_with_caps_lock);
 505                 list_del_init(&ci->i_snap_realm_item);
 506                 spin_unlock(&realm->inodes_with_caps_lock);
 507                 ceph_put_snap_realm(mdsc, realm);
 508         }
 509
 510         kfree(ci->i_symlink);
 511         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 512                 frag = rb_entry(n, struct ceph_inode_frag, node);
 513                 rb_erase(n, &ci->i_fragtree);
 514                 kfree(frag);
 515         }
 516
 517         __ceph_destroy_xattrs(ci);
 518         if (ci->i_xattrs.blob)
 519                 ceph_buffer_put(ci->i_xattrs.blob);
 520         if (ci->i_xattrs.prealloc_blob)
 521                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 522
 523         call_rcu(&inode->i_rcu, ceph_i_callback);
 524 }
 525
 526 int ceph_drop_inode(struct inode *inode)
 527 {
 528         /*
 529          * Positve dentry and corresponding inode are always accompanied
 530          * in MDS reply. So no need to keep inode in the cache after
 531          * dropping all its aliases.
 532          */
 533         return 1;
 534 }
 535
 536 /*
 537  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 538  * careful because either the client or MDS may have more up to date
 539  * info, depending on which capabilities are held, and whether
 540  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 541  * and size are monotonically increasing, except when utimes() or
 542  * truncate() increments the corresponding _seq values.)
 543  */
 544 int ceph_fill_file_size(struct inode *inode, int issued,
 545                         u32 truncate_seq, u64 truncate_size, u64 size)
 546 {
 547         struct ceph_inode_info *ci = ceph_inode(inode);
 548         int queue_trunc = 0;
 549
 550         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 551             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 552                 dout("size %lld -> %llu\n", inode->i_size, size);
 553                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 554                         pr_err("fill_file_size non-zero size for directory\n");
 555                         size = 0;
 556                 }
 557                 i_size_write(inode, size);
 558                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 559                 ci->i_reported_size = size;
 560                 if (truncate_seq != ci->i_truncate_seq) {
 561                         dout("truncate_seq %u -> %u\n",
 562                              ci->i_truncate_seq, truncate_seq);
 563                         ci->i_truncate_seq = truncate_seq;
 564
 565                         /* the MDS should have revoked these caps */
 566                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 567                                                CEPH_CAP_FILE_RD |
 568                                                CEPH_CAP_FILE_WR |
 569                                                CEPH_CAP_FILE_LAZYIO));
 570                         /*
 571                          * If we hold relevant caps, or in the case where we're
 572                          * not the only client referencing this file and we
 573                          * don't hold those caps, then we need to check whether
 574                          * the file is either opened or mmaped
 575                          */
 576                         if ((issued & (CEPH_CAP_FILE_CACHE|
 577                                        CEPH_CAP_FILE_BUFFER)) ||
 578                             mapping_mapped(inode->i_mapping) ||
 579                             __ceph_caps_file_wanted(ci)) {
 580                                 ci->i_truncate_pending++;
 581                                 queue_trunc = 1;
 582                         }
 583                 }
 584         }
 585         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 586             ci->i_truncate_size != truncate_size) {
 587                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 588                      truncate_size);
 589                 ci->i_truncate_size = truncate_size;
 590         }
 591
 592         if (queue_trunc)
 593                 ceph_fscache_invalidate(inode);
 594
 595         return queue_trunc;
 596 }
 597
 598 void ceph_fill_file_time(struct inode *inode, int issued,
 599                          u64 time_warp_seq, struct timespec *ctime,
 600                          struct timespec *mtime, struct timespec *atime)
 601 {
 602         struct ceph_inode_info *ci = ceph_inode(inode);
 603         int warn = 0;
 604
 605         if (issued & (CEPH_CAP_FILE_EXCL|
 606                       CEPH_CAP_FILE_WR|
 607                       CEPH_CAP_FILE_BUFFER|
 608                       CEPH_CAP_AUTH_EXCL|
 609                       CEPH_CAP_XATTR_EXCL)) {
 610                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 611                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 612                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 613                              ctime->tv_sec, ctime->tv_nsec);
 614                         inode->i_ctime = *ctime;
 615                 }
 616                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 617                         /* the MDS did a utimes() */
 618                         dout("mtime %ld.%09ld -> %ld.%09ld "
 619                              "tw %d -> %d\n",
 620                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 621                              mtime->tv_sec, mtime->tv_nsec,
 622                              ci->i_time_warp_seq, (int)time_warp_seq);
 623
 624                         inode->i_mtime = *mtime;
 625                         inode->i_atime = *atime;
 626                         ci->i_time_warp_seq = time_warp_seq;
 627                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 628                         /* nobody did utimes(); take the max */
 629                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 630                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 631                                      inode->i_mtime.tv_sec,
 632                                      inode->i_mtime.tv_nsec,
 633                                      mtime->tv_sec, mtime->tv_nsec);
 634                                 inode->i_mtime = *mtime;
 635                         }
 636                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 637                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 638                                      inode->i_atime.tv_sec,
 639                                      inode->i_atime.tv_nsec,
 640                                      atime->tv_sec, atime->tv_nsec);
 641                                 inode->i_atime = *atime;
 642                         }
 643                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 644                         /* we did a utimes(); ignore mds values */
 645                 } else {
 646                         warn = 1;
 647                 }
 648         } else {
 649                 /* we have no write|excl caps; whatever the MDS says is true */
 650                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 651                         inode->i_ctime = *ctime;
 652                         inode->i_mtime = *mtime;
 653                         inode->i_atime = *atime;
 654                         ci->i_time_warp_seq = time_warp_seq;
 655                 } else {
 656                         warn = 1;
 657                 }
 658         }
 659         if (warn) /* time_warp_seq shouldn't go backwards */
 660                 dout("%p mds time_warp_seq %llu < %u\n",
 661                      inode, time_warp_seq, ci->i_time_warp_seq);
 662 }
 663
 664 /*
 665  * Populate an inode based on info from mds.  May be called on new or
 666  * existing inodes.
 667  */
 668 static int fill_inode(struct inode *inode, struct page *locked_page,
 669                       struct ceph_mds_reply_info_in *iinfo,
 670                       struct ceph_mds_reply_dirfrag *dirinfo,
 671                       struct ceph_mds_session *session,
 672                       unsigned long ttl_from, int cap_fmode,
 673                       struct ceph_cap_reservation *caps_reservation)
 674 {
 675         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 676         struct ceph_mds_reply_inode *info = iinfo->in;
 677         struct ceph_inode_info *ci = ceph_inode(inode);
 678         int issued = 0, implemented, new_issued;
 679         struct timespec mtime, atime, ctime;
 680         struct ceph_buffer *xattr_blob = NULL;
 681         struct ceph_cap *new_cap = NULL;
 682         int err = 0;
 683         bool wake = false;
 684         bool queue_trunc = false;
 685         bool new_version = false;
 686         bool fill_inline = false;
 687
 688         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 689              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 690              ci->i_version);
 691
 692         /* prealloc new cap struct */
 693         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 694                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 695
 696         /*
 697          * prealloc xattr data, if it looks like we'll need it.  only
 698          * if len > 4 (meaning there are actually xattrs; the first 4
 699          * bytes are the xattr count).
 700          */
 701         if (iinfo->xattr_len > 4) {
 702                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 703                 if (!xattr_blob)
 704                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 705                                iinfo->xattr_len);
 706         }
 707
 708         spin_lock(&ci->i_ceph_lock);
 709
 710         /*
 711          * provided version will be odd if inode value is projected,
 712          * even if stable.  skip the update if we have newer stable
 713          * info (ours>=theirs, e.g. due to racing mds replies), unless
 714          * we are getting projected (unstable) info (in which case the
 715          * version is odd, and we want ours>theirs).
 716          *   us   them
 717          *   2    2     skip
 718          *   3    2     skip
 719          *   3    3     update
 720          */
 721         if (ci->i_version == 0 ||
 722             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 723              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 724                 new_version = true;
 725
 726         issued = __ceph_caps_issued(ci, &implemented);
 727         issued |= implemented | __ceph_caps_dirty(ci);
 728         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 729
 730         /* update inode */
 731         ci->i_version = le64_to_cpu(info->version);
 732         inode->i_version++;
 733         inode->i_rdev = le32_to_cpu(info->rdev);
 734         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 735
 736         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 737             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 738                 inode->i_mode = le32_to_cpu(info->mode);
 739                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 740                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 741                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 742                      from_kuid(&init_user_ns, inode->i_uid),
 743                      from_kgid(&init_user_ns, inode->i_gid));
 744         }
 745
 746         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 747             (issued & CEPH_CAP_LINK_EXCL) == 0)
 748                 set_nlink(inode, le32_to_cpu(info->nlink));
 749
 750         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 751                 /* be careful with mtime, atime, size */
 752                 ceph_decode_timespec(&atime, &info->atime);
 753                 ceph_decode_timespec(&mtime, &info->mtime);
 754                 ceph_decode_timespec(&ctime, &info->ctime);
 755                 ceph_fill_file_time(inode, issued,
 756                                 le32_to_cpu(info->time_warp_seq),
 757                                 &ctime, &mtime, &atime);
 758         }
 759
 760         if (new_version ||
 761             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 762                 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
 763                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 764                 ci->i_layout = info->layout;
 765                 ci->i_pool_ns_len = iinfo->pool_ns_len;
 766
 767                 queue_trunc = ceph_fill_file_size(inode, issued,
 768                                         le32_to_cpu(info->truncate_seq),
 769                                         le64_to_cpu(info->truncate_size),
 770                                         le64_to_cpu(info->size));
 771                 /* only update max_size on auth cap */
 772                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 773                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 774                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 775                                         le64_to_cpu(info->max_size));
 776                         ci->i_max_size = le64_to_cpu(info->max_size);
 777                 }
 778         }
 779
 780         /* xattrs */
 781         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 782         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 783             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 784                 if (ci->i_xattrs.blob)
 785                         ceph_buffer_put(ci->i_xattrs.blob);
 786                 ci->i_xattrs.blob = xattr_blob;
 787                 if (xattr_blob)
 788                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 789                                iinfo->xattr_data, iinfo->xattr_len);
 790                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 791                 ceph_forget_all_cached_acls(inode);
 792                 xattr_blob = NULL;
 793         }
 794
 795         inode->i_mapping->a_ops = &ceph_aops;
 796
 797         switch (inode->i_mode & S_IFMT) {
 798         case S_IFIFO:
 799         case S_IFBLK:
 800         case S_IFCHR:
 801         case S_IFSOCK:
 802                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 803                 inode->i_op = &ceph_file_iops;
 804                 break;
 805         case S_IFREG:
 806                 inode->i_op = &ceph_file_iops;
 807                 inode->i_fop = &ceph_file_fops;
 808                 break;
 809         case S_IFLNK:
 810                 inode->i_op = &ceph_symlink_iops;
 811                 if (!ci->i_symlink) {
 812                         u32 symlen = iinfo->symlink_len;
 813                         char *sym;
 814
 815                         spin_unlock(&ci->i_ceph_lock);
 816
 817                         err = -EINVAL;
 818                         if (WARN_ON(symlen != i_size_read(inode)))
 819                                 goto out;
 820
 821                         err = -ENOMEM;
 822                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 823                         if (!sym)
 824                                 goto out;
 825
 826                         spin_lock(&ci->i_ceph_lock);
 827                         if (!ci->i_symlink)
 828                                 ci->i_symlink = sym;
 829                         else
 830                                 kfree(sym); /* lost a race */
 831                 }
 832                 inode->i_link = ci->i_symlink;
 833                 break;
 834         case S_IFDIR:
 835                 inode->i_op = &ceph_dir_iops;
 836                 inode->i_fop = &ceph_dir_fops;
 837
 838                 ci->i_dir_layout = iinfo->dir_layout;
 839
 840                 ci->i_files = le64_to_cpu(info->files);
 841                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 842                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 843                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 844                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 845                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 846                 break;
 847         default:
 848                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 849                        ceph_vinop(inode), inode->i_mode);
 850         }
 851
 852         /* were we issued a capability? */
 853         if (info->cap.caps) {
 854                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 855                         unsigned caps = le32_to_cpu(info->cap.caps);
 856                         ceph_add_cap(inode, session,
 857                                      le64_to_cpu(info->cap.cap_id),
 858                                      cap_fmode, caps,
 859                                      le32_to_cpu(info->cap.wanted),
 860                                      le32_to_cpu(info->cap.seq),
 861                                      le32_to_cpu(info->cap.mseq),
 862                                      le64_to_cpu(info->cap.realm),
 863                                      info->cap.flags, &new_cap);
 864
 865                         /* set dir completion flag? */
 866                         if (S_ISDIR(inode->i_mode) &&
 867                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 868                             (caps & CEPH_CAP_FILE_SHARED) &&
 869                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 870                             !__ceph_dir_is_complete(ci)) {
 871                                 dout(" marking %p complete (empty)\n", inode);
 872                                 i_size_write(inode, 0);
 873                                 __ceph_dir_set_complete(ci,
 874                                         atomic64_read(&ci->i_release_count),
 875                                         atomic64_read(&ci->i_ordered_count));
 876                         }
 877
 878                         wake = true;
 879                 } else {
 880                         dout(" %p got snap_caps %s\n", inode,
 881                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 882                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 883                         if (cap_fmode >= 0)
 884                                 __ceph_get_fmode(ci, cap_fmode);
 885                 }
 886         } else if (cap_fmode >= 0) {
 887                 pr_warn("mds issued no caps on %llx.%llx\n",
 888                            ceph_vinop(inode));
 889                 __ceph_get_fmode(ci, cap_fmode);
 890         }
 891
 892         if (iinfo->inline_version > 0 &&
 893             iinfo->inline_version >= ci->i_inline_version) {
 894                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 895                 ci->i_inline_version = iinfo->inline_version;
 896                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 897                     (locked_page ||
 898                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 899                         fill_inline = true;
 900         }
 901
 902         spin_unlock(&ci->i_ceph_lock);
 903
 904         if (fill_inline)
 905                 ceph_fill_inline_data(inode, locked_page,
 906                                       iinfo->inline_data, iinfo->inline_len);
 907
 908         if (wake)
 909                 wake_up_all(&ci->i_cap_wq);
 910
 911         /* queue truncate if we saw i_size decrease */
 912         if (queue_trunc)
 913                 ceph_queue_vmtruncate(inode);
 914
 915         /* populate frag tree */
 916         if (S_ISDIR(inode->i_mode))
 917                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 918
 919         /* update delegation info? */
 920         if (dirinfo)
 921                 ceph_fill_dirfrag(inode, dirinfo);
 922
 923         err = 0;
 924 out:
 925         if (new_cap)
 926                 ceph_put_cap(mdsc, new_cap);
 927         if (xattr_blob)
 928                 ceph_buffer_put(xattr_blob);
 929         return err;
 930 }
 931
 932 /*
 933  * caller should hold session s_mutex.
 934  */
 935 static void update_dentry_lease(struct dentry *dentry,
 936                                 struct ceph_mds_reply_lease *lease,
 937                                 struct ceph_mds_session *session,
 938                                 unsigned long from_time)
 939 {
 940         struct ceph_dentry_info *di = ceph_dentry(dentry);
 941         long unsigned duration = le32_to_cpu(lease->duration_ms);
 942         long unsigned ttl = from_time + (duration * HZ) / 1000;
 943         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 944         struct inode *dir;
 945
 946         /* only track leases on regular dentries */
 947         if (dentry->d_op != &ceph_dentry_ops)
 948                 return;
 949
 950         spin_lock(&dentry->d_lock);
 951         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 952              dentry, duration, ttl);
 953
 954         /* make lease_rdcache_gen match directory */
 955         dir = d_inode(dentry->d_parent);
 956         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 957
 958         if (duration == 0)
 959                 goto out_unlock;
 960
 961         if (di->lease_gen == session->s_cap_gen &&
 962             time_before(ttl, dentry->d_time))
 963                 goto out_unlock;  /* we already have a newer lease. */
 964
 965         if (di->lease_session && di->lease_session != session)
 966                 goto out_unlock;
 967
 968         ceph_dentry_lru_touch(dentry);
 969
 970         if (!di->lease_session)
 971                 di->lease_session = ceph_get_mds_session(session);
 972         di->lease_gen = session->s_cap_gen;
 973         di->lease_seq = le32_to_cpu(lease->seq);
 974         di->lease_renew_after = half_ttl;
 975         di->lease_renew_from = 0;
 976         dentry->d_time = ttl;
 977 out_unlock:
 978         spin_unlock(&dentry->d_lock);
 979         return;
 980 }
 981
 982 /*
 983  * splice a dentry to an inode.
 984  * caller must hold directory i_mutex for this to be safe.
 985  */
 986 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
 987 {
 988         struct dentry *realdn;
 989
 990         BUG_ON(d_inode(dn));
 991
 992         /* dn must be unhashed */
 993         if (!d_unhashed(dn))
 994                 d_drop(dn);
 995         realdn = d_splice_alias(in, dn);
 996         if (IS_ERR(realdn)) {
 997                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 998                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 999                 dn = realdn; /* note realdn contains the error */
1000                 goto out;
1001         } else if (realdn) {
1002                 dout("dn %p (%d) spliced with %p (%d) "
1003                      "inode %p ino %llx.%llx\n",
1004                      dn, d_count(dn),
1005                      realdn, d_count(realdn),
1006                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1007                 dput(dn);
1008                 dn = realdn;
1009         } else {
1010                 BUG_ON(!ceph_dentry(dn));
1011                 dout("dn %p attached to %p ino %llx.%llx\n",
1012                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1013         }
1014 out:
1015         return dn;
1016 }
1017
1018 /*
1019  * Incorporate results into the local cache.  This is either just
1020  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1021  * after a lookup).
1022  *
1023  * A reply may contain
1024  *         a directory inode along with a dentry.
1025  *  and/or a target inode
1026  *
1027  * Called with snap_rwsem (read).
1028  */
1029 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1030                     struct ceph_mds_session *session)
1031 {
1032         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1033         struct inode *in = NULL;
1034         struct ceph_vino vino;
1035         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1036         int err = 0;
1037
1038         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1039              rinfo->head->is_dentry, rinfo->head->is_target);
1040
1041 #if 0
1042         /*
1043          * Debugging hook:
1044          *
1045          * If we resend completed ops to a recovering mds, we get no
1046          * trace.  Since that is very rare, pretend this is the case
1047          * to ensure the 'no trace' handlers in the callers behave.
1048          *
1049          * Fill in inodes unconditionally to avoid breaking cap
1050          * invariants.
1051          */
1052         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1053                 pr_info("fill_trace faking empty trace on %lld %s\n",
1054                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1055                 if (rinfo->head->is_dentry) {
1056                         rinfo->head->is_dentry = 0;
1057                         err = fill_inode(req->r_locked_dir,
1058                                          &rinfo->diri, rinfo->dirfrag,
1059                                          session, req->r_request_started, -1);
1060                 }
1061                 if (rinfo->head->is_target) {
1062                         rinfo->head->is_target = 0;
1063                         ininfo = rinfo->targeti.in;
1064                         vino.ino = le64_to_cpu(ininfo->ino);
1065                         vino.snap = le64_to_cpu(ininfo->snapid);
1066                         in = ceph_get_inode(sb, vino);
1067                         err = fill_inode(in, &rinfo->targeti, NULL,
1068                                          session, req->r_request_started,
1069                                          req->r_fmode);
1070                         iput(in);
1071                 }
1072         }
1073 #endif
1074
1075         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1076                 dout("fill_trace reply is empty!\n");
1077                 if (rinfo->head->result == 0 && req->r_locked_dir)
1078                         ceph_invalidate_dir_request(req);
1079                 return 0;
1080         }
1081
1082         if (rinfo->head->is_dentry) {
1083                 struct inode *dir = req->r_locked_dir;
1084
1085                 if (dir) {
1086                         err = fill_inode(dir, NULL,
1087                                          &rinfo->diri, rinfo->dirfrag,
1088                                          session, req->r_request_started, -1,
1089                                          &req->r_caps_reservation);
1090                         if (err < 0)
1091                                 goto done;
1092                 } else {
1093                         WARN_ON_ONCE(1);
1094                 }
1095
1096                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1097                         struct qstr dname;
1098                         struct dentry *dn, *parent;
1099
1100                         BUG_ON(!rinfo->head->is_target);
1101                         BUG_ON(req->r_dentry);
1102
1103                         parent = d_find_any_alias(dir);
1104                         BUG_ON(!parent);
1105
1106                         dname.name = rinfo->dname;
1107                         dname.len = rinfo->dname_len;
1108                         dname.hash = full_name_hash(dname.name, dname.len);
1109                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1110                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1111 retry_lookup:
1112                         dn = d_lookup(parent, &dname);
1113                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1114                              parent, dname.len, dname.name, dn);
1115
1116                         if (!dn) {
1117                                 dn = d_alloc(parent, &dname);
1118                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1119                                      dname.len, dname.name, dn);
1120                                 if (dn == NULL) {
1121                                         dput(parent);
1122                                         err = -ENOMEM;
1123                                         goto done;
1124                                 }
1125                                 err = ceph_init_dentry(dn);
1126                                 if (err < 0) {
1127                                         dput(dn);
1128                                         dput(parent);
1129                                         goto done;
1130                                 }
1131                         } else if (d_really_is_positive(dn) &&
1132                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1133                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1134                                 dout(" dn %p points to wrong inode %p\n",
1135                                      dn, d_inode(dn));
1136                                 d_delete(dn);
1137                                 dput(dn);
1138                                 goto retry_lookup;
1139                         }
1140
1141                         req->r_dentry = dn;
1142                         dput(parent);
1143                 }
1144         }
1145
1146         if (rinfo->head->is_target) {
1147                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1148                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1149
1150                 in = ceph_get_inode(sb, vino);
1151                 if (IS_ERR(in)) {
1152                         err = PTR_ERR(in);
1153                         goto done;
1154                 }
1155                 req->r_target_inode = in;
1156
1157                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1158                                 session, req->r_request_started,
1159                                 (!req->r_aborted && rinfo->head->result == 0) ?
1160                                 req->r_fmode : -1,
1161                                 &req->r_caps_reservation);
1162                 if (err < 0) {
1163                         pr_err("fill_inode badness %p %llx.%llx\n",
1164                                 in, ceph_vinop(in));
1165                         goto done;
1166                 }
1167         }
1168
1169         /*
1170          * ignore null lease/binding on snapdir ENOENT, or else we
1171          * will have trouble splicing in the virtual snapdir later
1172          */
1173         if (rinfo->head->is_dentry && !req->r_aborted &&
1174             req->r_locked_dir &&
1175             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1176                                                fsc->mount_options->snapdir_name,
1177                                                req->r_dentry->d_name.len))) {
1178                 /*
1179                  * lookup link rename   : null -> possibly existing inode
1180                  * mknod symlink mkdir  : null -> new inode
1181                  * unlink               : linked -> null
1182                  */
1183                 struct inode *dir = req->r_locked_dir;
1184                 struct dentry *dn = req->r_dentry;
1185                 bool have_dir_cap, have_lease;
1186
1187                 BUG_ON(!dn);
1188                 BUG_ON(!dir);
1189                 BUG_ON(d_inode(dn->d_parent) != dir);
1190                 BUG_ON(ceph_ino(dir) !=
1191                        le64_to_cpu(rinfo->diri.in->ino));
1192                 BUG_ON(ceph_snap(dir) !=
1193                        le64_to_cpu(rinfo->diri.in->snapid));
1194
1195                 /* do we have a lease on the whole dir? */
1196                 have_dir_cap =
1197                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1198                          CEPH_CAP_FILE_SHARED);
1199
1200                 /* do we have a dn lease? */
1201                 have_lease = have_dir_cap ||
1202                         le32_to_cpu(rinfo->dlease->duration_ms);
1203                 if (!have_lease)
1204                         dout("fill_trace  no dentry lease or dir cap\n");
1205
1206                 /* rename? */
1207                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1208                         struct inode *olddir = req->r_old_dentry_dir;
1209                         BUG_ON(!olddir);
1210
1211                         dout(" src %p '%pd' dst %p '%pd'\n",
1212                              req->r_old_dentry,
1213                              req->r_old_dentry,
1214                              dn, dn);
1215                         dout("fill_trace doing d_move %p -> %p\n",
1216                              req->r_old_dentry, dn);
1217
1218                         /* d_move screws up sibling dentries' offsets */
1219                         ceph_dir_clear_ordered(dir);
1220                         ceph_dir_clear_ordered(olddir);
1221
1222                         d_move(req->r_old_dentry, dn);
1223                         dout(" src %p '%pd' dst %p '%pd'\n",
1224                              req->r_old_dentry,
1225                              req->r_old_dentry,
1226                              dn, dn);
1227
1228                         /* ensure target dentry is invalidated, despite
1229                            rehashing bug in vfs_rename_dir */
1230                         ceph_invalidate_dentry_lease(dn);
1231
1232                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1233                              ceph_dentry(req->r_old_dentry)->offset);
1234
1235                         dn = req->r_old_dentry;  /* use old_dentry */
1236                 }
1237
1238                 /* null dentry? */
1239                 if (!rinfo->head->is_target) {
1240                         dout("fill_trace null dentry\n");
1241                         if (d_really_is_positive(dn)) {
1242                                 ceph_dir_clear_ordered(dir);
1243                                 dout("d_delete %p\n", dn);
1244                                 d_delete(dn);
1245                         } else {
1246                                 if (have_lease && d_unhashed(dn))
1247                                         d_add(dn, NULL);
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);
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                         d_invalidate(dn);
1270                         have_lease = false;
1271                 }
1272
1273                 if (have_lease)
1274                         update_dentry_lease(dn, rinfo->dlease, session,
1275                                             req->r_request_started);
1276                 dout(" final dn %p\n", dn);
1277         } else if (!req->r_aborted &&
1278                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1279                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1280                 struct dentry *dn = req->r_dentry;
1281                 struct inode *dir = req->r_locked_dir;
1282
1283                 /* fill out a snapdir LOOKUPSNAP dentry */
1284                 BUG_ON(!dn);
1285                 BUG_ON(!dir);
1286                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1287                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1288                 ceph_dir_clear_ordered(dir);
1289                 ihold(in);
1290                 dn = splice_dentry(dn, in);
1291                 if (IS_ERR(dn)) {
1292                         err = PTR_ERR(dn);
1293                         goto done;
1294                 }
1295                 req->r_dentry = dn;  /* may have spliced */
1296         }
1297 done:
1298         dout("fill_trace done err=%d\n", err);
1299         return err;
1300 }
1301
1302 /*
1303  * Prepopulate our cache with readdir results, leases, etc.
1304  */
1305 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1306                                            struct ceph_mds_session *session)
1307 {
1308         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1309         int i, err = 0;
1310
1311         for (i = 0; i < rinfo->dir_nr; i++) {
1312                 struct ceph_vino vino;
1313                 struct inode *in;
1314                 int rc;
1315
1316                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1317                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1318
1319                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1320                 if (IS_ERR(in)) {
1321                         err = PTR_ERR(in);
1322                         dout("new_inode badness got %d\n", err);
1323                         continue;
1324                 }
1325                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1326                                 req->r_request_started, -1,
1327                                 &req->r_caps_reservation);
1328                 if (rc < 0) {
1329                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1330                         err = rc;
1331                         continue;
1332                 }
1333         }
1334
1335         return err;
1336 }
1337
1338 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1339 {
1340         if (ctl->page) {
1341                 kunmap(ctl->page);
1342                 put_page(ctl->page);
1343                 ctl->page = NULL;
1344         }
1345 }
1346
1347 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1348                               struct ceph_readdir_cache_control *ctl,
1349                               struct ceph_mds_request *req)
1350 {
1351         struct ceph_inode_info *ci = ceph_inode(dir);
1352         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1353         unsigned idx = ctl->index % nsize;
1354         pgoff_t pgoff = ctl->index / nsize;
1355
1356         if (!ctl->page || pgoff != page_index(ctl->page)) {
1357                 ceph_readdir_cache_release(ctl);
1358                 if (idx == 0)
1359                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1360                 else
1361                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1362                 if (!ctl->page) {
1363                         ctl->index = -1;
1364                         return idx == 0 ? -ENOMEM : 0;
1365                 }
1366                 /* reading/filling the cache are serialized by
1367                  * i_mutex, no need to use page lock */
1368                 unlock_page(ctl->page);
1369                 ctl->dentries = kmap(ctl->page);
1370                 if (idx == 0)
1371                         memset(ctl->dentries, 0, PAGE_SIZE);
1372         }
1373
1374         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1375             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1376                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1377                 ctl->dentries[idx] = dn;
1378                 ctl->index++;
1379         } else {
1380                 dout("disable readdir cache\n");
1381                 ctl->index = -1;
1382         }
1383         return 0;
1384 }
1385
1386 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1387                              struct ceph_mds_session *session)
1388 {
1389         struct dentry *parent = req->r_dentry;
1390         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1391         struct qstr dname;
1392         struct dentry *dn;
1393         struct inode *in;
1394         int err = 0, skipped = 0, ret, i;
1395         struct inode *snapdir = NULL;
1396         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1397         struct ceph_dentry_info *di;
1398         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1399         struct ceph_readdir_cache_control cache_ctl = {};
1400
1401         if (req->r_aborted)
1402                 return readdir_prepopulate_inodes_only(req, session);
1403
1404         if (rinfo->dir_dir &&
1405             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1406                 dout("readdir_prepopulate got new frag %x -> %x\n",
1407                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1408                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1409                 if (ceph_frag_is_leftmost(frag))
1410                         req->r_readdir_offset = 2;
1411                 else
1412                         req->r_readdir_offset = 0;
1413         }
1414
1415         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1416                 snapdir = ceph_get_snapdir(d_inode(parent));
1417                 parent = d_find_alias(snapdir);
1418                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1419                      rinfo->dir_nr, parent);
1420         } else {
1421                 dout("readdir_prepopulate %d items under dn %p\n",
1422                      rinfo->dir_nr, parent);
1423                 if (rinfo->dir_dir)
1424                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1425         }
1426
1427         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1428                 /* note dir version at start of readdir so we can tell
1429                  * if any dentries get dropped */
1430                 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1431                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1432                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1433                 req->r_readdir_cache_idx = 0;
1434         }
1435
1436         cache_ctl.index = req->r_readdir_cache_idx;
1437
1438         /* FIXME: release caps/leases if error occurs */
1439         for (i = 0; i < rinfo->dir_nr; i++) {
1440                 struct ceph_vino vino;
1441
1442                 dname.name = rinfo->dir_dname[i];
1443                 dname.len = rinfo->dir_dname_len[i];
1444                 dname.hash = full_name_hash(dname.name, dname.len);
1445
1446                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1447                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1448
1449 retry_lookup:
1450                 dn = d_lookup(parent, &dname);
1451                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1452                      parent, dname.len, dname.name, dn);
1453
1454                 if (!dn) {
1455                         dn = d_alloc(parent, &dname);
1456                         dout("d_alloc %p '%.*s' = %p\n", parent,
1457                              dname.len, dname.name, dn);
1458                         if (dn == NULL) {
1459                                 dout("d_alloc badness\n");
1460                                 err = -ENOMEM;
1461                                 goto out;
1462                         }
1463                         ret = ceph_init_dentry(dn);
1464                         if (ret < 0) {
1465                                 dput(dn);
1466                                 err = ret;
1467                                 goto out;
1468                         }
1469                 } else if (d_really_is_positive(dn) &&
1470                            (ceph_ino(d_inode(dn)) != vino.ino ||
1471                             ceph_snap(d_inode(dn)) != vino.snap)) {
1472                         dout(" dn %p points to wrong inode %p\n",
1473                              dn, d_inode(dn));
1474                         d_delete(dn);
1475                         dput(dn);
1476                         goto retry_lookup;
1477                 }
1478
1479                 /* inode */
1480                 if (d_really_is_positive(dn)) {
1481                         in = d_inode(dn);
1482                 } else {
1483                         in = ceph_get_inode(parent->d_sb, vino);
1484                         if (IS_ERR(in)) {
1485                                 dout("new_inode badness\n");
1486                                 d_drop(dn);
1487                                 dput(dn);
1488                                 err = PTR_ERR(in);
1489                                 goto out;
1490                         }
1491                 }
1492
1493                 ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1494                                  req->r_request_started, -1,
1495                                  &req->r_caps_reservation);
1496                 if (ret < 0) {
1497                         pr_err("fill_inode badness on %p\n", in);
1498                         if (d_really_is_negative(dn))
1499                                 iput(in);
1500                         d_drop(dn);
1501                         err = ret;
1502                         goto next_item;
1503                 }
1504
1505                 if (d_really_is_negative(dn)) {
1506                         struct dentry *realdn;
1507
1508                         if (ceph_security_xattr_deadlock(in)) {
1509                                 dout(" skip splicing dn %p to inode %p"
1510                                      " (security xattr deadlock)\n", dn, in);
1511                                 iput(in);
1512                                 skipped++;
1513                                 goto next_item;
1514                         }
1515
1516                         realdn = splice_dentry(dn, in);
1517                         if (IS_ERR(realdn)) {
1518                                 err = PTR_ERR(realdn);
1519                                 d_drop(dn);
1520                                 dn = NULL;
1521                                 goto next_item;
1522                         }
1523                         dn = realdn;
1524                 }
1525
1526                 di = dn->d_fsdata;
1527                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1528
1529                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1530                                     req->r_session,
1531                                     req->r_request_started);
1532
1533                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1534                         ret = fill_readdir_cache(d_inode(parent), dn,
1535                                                  &cache_ctl, req);
1536                         if (ret < 0)
1537                                 err = ret;
1538                 }
1539 next_item:
1540                 if (dn)
1541                         dput(dn);
1542         }
1543 out:
1544         if (err == 0 && skipped == 0) {
1545                 req->r_did_prepopulate = true;
1546                 req->r_readdir_cache_idx = cache_ctl.index;
1547         }
1548         ceph_readdir_cache_release(&cache_ctl);
1549         if (snapdir) {
1550                 iput(snapdir);
1551                 dput(parent);
1552         }
1553         dout("readdir_prepopulate done\n");
1554         return err;
1555 }
1556
1557 int ceph_inode_set_size(struct inode *inode, loff_t size)
1558 {
1559         struct ceph_inode_info *ci = ceph_inode(inode);
1560         int ret = 0;
1561
1562         spin_lock(&ci->i_ceph_lock);
1563         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1564         i_size_write(inode, size);
1565         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1566
1567         /* tell the MDS if we are approaching max_size */
1568         if ((size << 1) >= ci->i_max_size &&
1569             (ci->i_reported_size << 1) < ci->i_max_size)
1570                 ret = 1;
1571
1572         spin_unlock(&ci->i_ceph_lock);
1573         return ret;
1574 }
1575
1576 /*
1577  * Write back inode data in a worker thread.  (This can't be done
1578  * in the message handler context.)
1579  */
1580 void ceph_queue_writeback(struct inode *inode)
1581 {
1582         ihold(inode);
1583         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1584                        &ceph_inode(inode)->i_wb_work)) {
1585                 dout("ceph_queue_writeback %p\n", inode);
1586         } else {
1587                 dout("ceph_queue_writeback %p failed\n", inode);
1588                 iput(inode);
1589         }
1590 }
1591
1592 static void ceph_writeback_work(struct work_struct *work)
1593 {
1594         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1595                                                   i_wb_work);
1596         struct inode *inode = &ci->vfs_inode;
1597
1598         dout("writeback %p\n", inode);
1599         filemap_fdatawrite(&inode->i_data);
1600         iput(inode);
1601 }
1602
1603 /*
1604  * queue an async invalidation
1605  */
1606 void ceph_queue_invalidate(struct inode *inode)
1607 {
1608         ihold(inode);
1609         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1610                        &ceph_inode(inode)->i_pg_inv_work)) {
1611                 dout("ceph_queue_invalidate %p\n", inode);
1612         } else {
1613                 dout("ceph_queue_invalidate %p failed\n", inode);
1614                 iput(inode);
1615         }
1616 }
1617
1618 /*
1619  * Invalidate inode pages in a worker thread.  (This can't be done
1620  * in the message handler context.)
1621  */
1622 static void ceph_invalidate_work(struct work_struct *work)
1623 {
1624         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1625                                                   i_pg_inv_work);
1626         struct inode *inode = &ci->vfs_inode;
1627         u32 orig_gen;
1628         int check = 0;
1629
1630         mutex_lock(&ci->i_truncate_mutex);
1631         spin_lock(&ci->i_ceph_lock);
1632         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1633              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1634         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1635                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1636                         check = 1;
1637                 spin_unlock(&ci->i_ceph_lock);
1638                 mutex_unlock(&ci->i_truncate_mutex);
1639                 goto out;
1640         }
1641         orig_gen = ci->i_rdcache_gen;
1642         spin_unlock(&ci->i_ceph_lock);
1643
1644         truncate_pagecache(inode, 0);
1645
1646         spin_lock(&ci->i_ceph_lock);
1647         if (orig_gen == ci->i_rdcache_gen &&
1648             orig_gen == ci->i_rdcache_revoking) {
1649                 dout("invalidate_pages %p gen %d successful\n", inode,
1650                      ci->i_rdcache_gen);
1651                 ci->i_rdcache_revoking--;
1652                 check = 1;
1653         } else {
1654                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1655                      inode, orig_gen, ci->i_rdcache_gen,
1656                      ci->i_rdcache_revoking);
1657                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1658                         check = 1;
1659         }
1660         spin_unlock(&ci->i_ceph_lock);
1661         mutex_unlock(&ci->i_truncate_mutex);
1662 out:
1663         if (check)
1664                 ceph_check_caps(ci, 0, NULL);
1665         iput(inode);
1666 }
1667
1668
1669 /*
1670  * called by trunc_wq;
1671  *
1672  * We also truncate in a separate thread as well.
1673  */
1674 static void ceph_vmtruncate_work(struct work_struct *work)
1675 {
1676         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1677                                                   i_vmtruncate_work);
1678         struct inode *inode = &ci->vfs_inode;
1679
1680         dout("vmtruncate_work %p\n", inode);
1681         __ceph_do_pending_vmtruncate(inode);
1682         iput(inode);
1683 }
1684
1685 /*
1686  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1687  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1688  */
1689 void ceph_queue_vmtruncate(struct inode *inode)
1690 {
1691         struct ceph_inode_info *ci = ceph_inode(inode);
1692
1693         ihold(inode);
1694
1695         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1696                        &ci->i_vmtruncate_work)) {
1697                 dout("ceph_queue_vmtruncate %p\n", inode);
1698         } else {
1699                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1700                      inode, ci->i_truncate_pending);
1701                 iput(inode);
1702         }
1703 }
1704
1705 /*
1706  * Make sure any pending truncation is applied before doing anything
1707  * that may depend on it.
1708  */
1709 void __ceph_do_pending_vmtruncate(struct inode *inode)
1710 {
1711         struct ceph_inode_info *ci = ceph_inode(inode);
1712         u64 to;
1713         int wrbuffer_refs, finish = 0;
1714
1715         mutex_lock(&ci->i_truncate_mutex);
1716 retry:
1717         spin_lock(&ci->i_ceph_lock);
1718         if (ci->i_truncate_pending == 0) {
1719                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1720                 spin_unlock(&ci->i_ceph_lock);
1721                 mutex_unlock(&ci->i_truncate_mutex);
1722                 return;
1723         }
1724
1725         /*
1726          * make sure any dirty snapped pages are flushed before we
1727          * possibly truncate them.. so write AND block!
1728          */
1729         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1730                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1731                      inode);
1732                 spin_unlock(&ci->i_ceph_lock);
1733                 filemap_write_and_wait_range(&inode->i_data, 0,
1734                                              inode->i_sb->s_maxbytes);
1735                 goto retry;
1736         }
1737
1738         /* there should be no reader or writer */
1739         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1740
1741         to = ci->i_truncate_size;
1742         wrbuffer_refs = ci->i_wrbuffer_ref;
1743         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1744              ci->i_truncate_pending, to);
1745         spin_unlock(&ci->i_ceph_lock);
1746
1747         truncate_pagecache(inode, to);
1748
1749         spin_lock(&ci->i_ceph_lock);
1750         if (to == ci->i_truncate_size) {
1751                 ci->i_truncate_pending = 0;
1752                 finish = 1;
1753         }
1754         spin_unlock(&ci->i_ceph_lock);
1755         if (!finish)
1756                 goto retry;
1757
1758         mutex_unlock(&ci->i_truncate_mutex);
1759
1760         if (wrbuffer_refs == 0)
1761                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1762
1763         wake_up_all(&ci->i_cap_wq);
1764 }
1765
1766 /*
1767  * symlinks
1768  */
1769 static const struct inode_operations ceph_symlink_iops = {
1770         .readlink = generic_readlink,
1771         .get_link = simple_get_link,
1772         .setattr = ceph_setattr,
1773         .getattr = ceph_getattr,
1774         .setxattr = generic_setxattr,
1775         .getxattr = generic_getxattr,
1776         .listxattr = ceph_listxattr,
1777         .removexattr = generic_removexattr,
1778 };
1779
1780 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1781 {
1782         struct ceph_inode_info *ci = ceph_inode(inode);
1783         const unsigned int ia_valid = attr->ia_valid;
1784         struct ceph_mds_request *req;
1785         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1786         struct ceph_cap_flush *prealloc_cf;
1787         int issued;
1788         int release = 0, dirtied = 0;
1789         int mask = 0;
1790         int err = 0;
1791         int inode_dirty_flags = 0;
1792         bool lock_snap_rwsem = false;
1793
1794         if (ceph_snap(inode) != CEPH_NOSNAP)
1795                 return -EROFS;
1796
1797         err = inode_change_ok(inode, attr);
1798         if (err != 0)
1799                 return err;
1800
1801         prealloc_cf = ceph_alloc_cap_flush();
1802         if (!prealloc_cf)
1803                 return -ENOMEM;
1804
1805         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1806                                        USE_AUTH_MDS);
1807         if (IS_ERR(req)) {
1808                 ceph_free_cap_flush(prealloc_cf);
1809                 return PTR_ERR(req);
1810         }
1811
1812         spin_lock(&ci->i_ceph_lock);
1813         issued = __ceph_caps_issued(ci, NULL);
1814
1815         if (!ci->i_head_snapc &&
1816             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1817                 lock_snap_rwsem = true;
1818                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1819                         spin_unlock(&ci->i_ceph_lock);
1820                         down_read(&mdsc->snap_rwsem);
1821                         spin_lock(&ci->i_ceph_lock);
1822                         issued = __ceph_caps_issued(ci, NULL);
1823                 }
1824         }
1825
1826         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1827
1828         if (ia_valid & ATTR_UID) {
1829                 dout("setattr %p uid %d -> %d\n", inode,
1830                      from_kuid(&init_user_ns, inode->i_uid),
1831                      from_kuid(&init_user_ns, attr->ia_uid));
1832                 if (issued & CEPH_CAP_AUTH_EXCL) {
1833                         inode->i_uid = attr->ia_uid;
1834                         dirtied |= CEPH_CAP_AUTH_EXCL;
1835                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1836                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1837                         req->r_args.setattr.uid = cpu_to_le32(
1838                                 from_kuid(&init_user_ns, attr->ia_uid));
1839                         mask |= CEPH_SETATTR_UID;
1840                         release |= CEPH_CAP_AUTH_SHARED;
1841                 }
1842         }
1843         if (ia_valid & ATTR_GID) {
1844                 dout("setattr %p gid %d -> %d\n", inode,
1845                      from_kgid(&init_user_ns, inode->i_gid),
1846                      from_kgid(&init_user_ns, attr->ia_gid));
1847                 if (issued & CEPH_CAP_AUTH_EXCL) {
1848                         inode->i_gid = attr->ia_gid;
1849                         dirtied |= CEPH_CAP_AUTH_EXCL;
1850                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1851                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1852                         req->r_args.setattr.gid = cpu_to_le32(
1853                                 from_kgid(&init_user_ns, attr->ia_gid));
1854                         mask |= CEPH_SETATTR_GID;
1855                         release |= CEPH_CAP_AUTH_SHARED;
1856                 }
1857         }
1858         if (ia_valid & ATTR_MODE) {
1859                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1860                      attr->ia_mode);
1861                 if (issued & CEPH_CAP_AUTH_EXCL) {
1862                         inode->i_mode = attr->ia_mode;
1863                         dirtied |= CEPH_CAP_AUTH_EXCL;
1864                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1865                            attr->ia_mode != inode->i_mode) {
1866                         inode->i_mode = attr->ia_mode;
1867                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1868                         mask |= CEPH_SETATTR_MODE;
1869                         release |= CEPH_CAP_AUTH_SHARED;
1870                 }
1871         }
1872
1873         if (ia_valid & ATTR_ATIME) {
1874                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1875                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1876                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1877                 if (issued & CEPH_CAP_FILE_EXCL) {
1878                         ci->i_time_warp_seq++;
1879                         inode->i_atime = attr->ia_atime;
1880                         dirtied |= CEPH_CAP_FILE_EXCL;
1881                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1882                            timespec_compare(&inode->i_atime,
1883                                             &attr->ia_atime) < 0) {
1884                         inode->i_atime = attr->ia_atime;
1885                         dirtied |= CEPH_CAP_FILE_WR;
1886                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1887                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1888                         ceph_encode_timespec(&req->r_args.setattr.atime,
1889                                              &attr->ia_atime);
1890                         mask |= CEPH_SETATTR_ATIME;
1891                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1892                                 CEPH_CAP_FILE_WR;
1893                 }
1894         }
1895         if (ia_valid & ATTR_MTIME) {
1896                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1897                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1898                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1899                 if (issued & CEPH_CAP_FILE_EXCL) {
1900                         ci->i_time_warp_seq++;
1901                         inode->i_mtime = attr->ia_mtime;
1902                         dirtied |= CEPH_CAP_FILE_EXCL;
1903                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1904                            timespec_compare(&inode->i_mtime,
1905                                             &attr->ia_mtime) < 0) {
1906                         inode->i_mtime = attr->ia_mtime;
1907                         dirtied |= CEPH_CAP_FILE_WR;
1908                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1909                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1910                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1911                                              &attr->ia_mtime);
1912                         mask |= CEPH_SETATTR_MTIME;
1913                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1914                                 CEPH_CAP_FILE_WR;
1915                 }
1916         }
1917         if (ia_valid & ATTR_SIZE) {
1918                 dout("setattr %p size %lld -> %lld\n", inode,
1919                      inode->i_size, attr->ia_size);
1920                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1921                     attr->ia_size > inode->i_size) {
1922                         i_size_write(inode, attr->ia_size);
1923                         inode->i_blocks =
1924                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1925                         inode->i_ctime = attr->ia_ctime;
1926                         ci->i_reported_size = attr->ia_size;
1927                         dirtied |= CEPH_CAP_FILE_EXCL;
1928                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1929                            attr->ia_size != inode->i_size) {
1930                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1931                         req->r_args.setattr.old_size =
1932                                 cpu_to_le64(inode->i_size);
1933                         mask |= CEPH_SETATTR_SIZE;
1934                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1935                                 CEPH_CAP_FILE_WR;
1936                 }
1937         }
1938
1939         /* these do nothing */
1940         if (ia_valid & ATTR_CTIME) {
1941                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1942                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1943                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1944                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1945                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1946                      only ? "ctime only" : "ignored");
1947                 inode->i_ctime = attr->ia_ctime;
1948                 if (only) {
1949                         /*
1950                          * if kernel wants to dirty ctime but nothing else,
1951                          * we need to choose a cap to dirty under, or do
1952                          * a almost-no-op setattr
1953                          */
1954                         if (issued & CEPH_CAP_AUTH_EXCL)
1955                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1956                         else if (issued & CEPH_CAP_FILE_EXCL)
1957                                 dirtied |= CEPH_CAP_FILE_EXCL;
1958                         else if (issued & CEPH_CAP_XATTR_EXCL)
1959                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1960                         else
1961                                 mask |= CEPH_SETATTR_CTIME;
1962                 }
1963         }
1964         if (ia_valid & ATTR_FILE)
1965                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1966
1967         if (dirtied) {
1968                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
1969                                                            &prealloc_cf);
1970                 inode->i_ctime = current_fs_time(inode->i_sb);
1971         }
1972
1973         release &= issued;
1974         spin_unlock(&ci->i_ceph_lock);
1975         if (lock_snap_rwsem)
1976                 up_read(&mdsc->snap_rwsem);
1977
1978         if (inode_dirty_flags)
1979                 __mark_inode_dirty(inode, inode_dirty_flags);
1980
1981         if (ia_valid & ATTR_MODE) {
1982                 err = posix_acl_chmod(inode, attr->ia_mode);
1983                 if (err)
1984                         goto out_put;
1985         }
1986
1987         if (mask) {
1988                 req->r_inode = inode;
1989                 ihold(inode);
1990                 req->r_inode_drop = release;
1991                 req->r_args.setattr.mask = cpu_to_le32(mask);
1992                 req->r_num_caps = 1;
1993                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1994         }
1995         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1996              ceph_cap_string(dirtied), mask);
1997
1998         ceph_mdsc_put_request(req);
1999         if (mask & CEPH_SETATTR_SIZE)
2000                 __ceph_do_pending_vmtruncate(inode);
2001         ceph_free_cap_flush(prealloc_cf);
2002         return err;
2003 out_put:
2004         ceph_mdsc_put_request(req);
2005         ceph_free_cap_flush(prealloc_cf);
2006         return err;
2007 }
2008
2009 /*
2010  * setattr
2011  */
2012 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2013 {
2014         return __ceph_setattr(d_inode(dentry), attr);
2015 }
2016
2017 /*
2018  * Verify that we have a lease on the given mask.  If not,
2019  * do a getattr against an mds.
2020  */
2021 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2022                       int mask, bool force)
2023 {
2024         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2025         struct ceph_mds_client *mdsc = fsc->mdsc;
2026         struct ceph_mds_request *req;
2027         int err;
2028
2029         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2030                 dout("do_getattr inode %p SNAPDIR\n", inode);
2031                 return 0;
2032         }
2033
2034         dout("do_getattr inode %p mask %s mode 0%o\n",
2035              inode, ceph_cap_string(mask), inode->i_mode);
2036         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2037                 return 0;
2038
2039         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2040         if (IS_ERR(req))
2041                 return PTR_ERR(req);
2042         req->r_inode = inode;
2043         ihold(inode);
2044         req->r_num_caps = 1;
2045         req->r_args.getattr.mask = cpu_to_le32(mask);
2046         req->r_locked_page = locked_page;
2047         err = ceph_mdsc_do_request(mdsc, NULL, req);
2048         if (locked_page && err == 0) {
2049                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2050                 if (inline_version == 0) {
2051                         /* the reply is supposed to contain inline data */
2052                         err = -EINVAL;
2053                 } else if (inline_version == CEPH_INLINE_NONE) {
2054                         err = -ENODATA;
2055                 } else {
2056                         err = req->r_reply_info.targeti.inline_len;
2057                 }
2058         }
2059         ceph_mdsc_put_request(req);
2060         dout("do_getattr result=%d\n", err);
2061         return err;
2062 }
2063
2064
2065 /*
2066  * Check inode permissions.  We verify we have a valid value for
2067  * the AUTH cap, then call the generic handler.
2068  */
2069 int ceph_permission(struct inode *inode, int mask)
2070 {
2071         int err;
2072
2073         if (mask & MAY_NOT_BLOCK)
2074                 return -ECHILD;
2075
2076         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2077
2078         if (!err)
2079                 err = generic_permission(inode, mask);
2080         return err;
2081 }
2082
2083 /*
2084  * Get all attributes.  Hopefully somedata we'll have a statlite()
2085  * and can limit the fields we require to be accurate.
2086  */
2087 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2088                  struct kstat *stat)
2089 {
2090         struct inode *inode = d_inode(dentry);
2091         struct ceph_inode_info *ci = ceph_inode(inode);
2092         int err;
2093
2094         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2095         if (!err) {
2096                 generic_fillattr(inode, stat);
2097                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2098                 if (ceph_snap(inode) != CEPH_NOSNAP)
2099                         stat->dev = ceph_snap(inode);
2100                 else
2101                         stat->dev = 0;
2102                 if (S_ISDIR(inode->i_mode)) {
2103                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2104                                                 RBYTES))
2105                                 stat->size = ci->i_rbytes;
2106                         else
2107                                 stat->size = ci->i_files + ci->i_subdirs;
2108                         stat->blocks = 0;
2109                         stat->blksize = 65536;
2110                 }
2111         }
2112         return err;
2113 }