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 #include <linux/sort.h>
  15
  16 #include "super.h"
  17 #include "mds_client.h"
  18 #include "cache.h"
  19 #include <linux/ceph/decode.h>
  20
  21 /*
  22  * Ceph inode operations
  23  *
  24  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  25  * setattr, etc.), xattr helpers, and helpers for assimilating
  26  * metadata returned by the MDS into our cache.
  27  *
  28  * Also define helpers for doing asynchronous writeback, invalidation,
  29  * and truncation for the benefit of those who can't afford to block
  30  * (typically because they are in the message handler path).
  31  */
  32
  33 static const struct inode_operations ceph_symlink_iops;
  34
  35 static void ceph_invalidate_work(struct work_struct *work);
  36 static void ceph_writeback_work(struct work_struct *work);
  37 static void ceph_vmtruncate_work(struct work_struct *work);
  38
  39 /*
  40  * find or create an inode, given the ceph ino number
  41  */
  42 static int ceph_set_ino_cb(struct inode *inode, void *data)
  43 {
  44         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  45         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  46         return 0;
  47 }
  48
  49 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  50 {
  51         struct inode *inode;
  52         ino_t t = ceph_vino_to_ino(vino);
  53
  54         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  55         if (inode == NULL)
  56                 return ERR_PTR(-ENOMEM);
  57         if (inode->i_state & I_NEW) {
  58                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  59                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  60                 unlock_new_inode(inode);
  61         }
  62
  63         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  64              vino.snap, inode);
  65         return inode;
  66 }
  67
  68 /*
  69  * get/constuct snapdir inode for a given directory
  70  */
  71 struct inode *ceph_get_snapdir(struct inode *parent)
  72 {
  73         struct ceph_vino vino = {
  74                 .ino = ceph_ino(parent),
  75                 .snap = CEPH_SNAPDIR,
  76         };
  77         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  78         struct ceph_inode_info *ci = ceph_inode(inode);
  79
  80         BUG_ON(!S_ISDIR(parent->i_mode));
  81         if (IS_ERR(inode))
  82                 return inode;
  83         inode->i_mode = parent->i_mode;
  84         inode->i_uid = parent->i_uid;
  85         inode->i_gid = parent->i_gid;
  86         inode->i_op = &ceph_snapdir_iops;
  87         inode->i_fop = &ceph_snapdir_fops;
  88         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  89         ci->i_rbytes = 0;
  90         return inode;
  91 }
  92
  93 const struct inode_operations ceph_file_iops = {
  94         .permission = ceph_permission,
  95         .setattr = ceph_setattr,
  96         .getattr = ceph_getattr,
  97         .setxattr = generic_setxattr,
  98         .getxattr = generic_getxattr,
  99         .listxattr = ceph_listxattr,
 100         .removexattr = generic_removexattr,
 101         .get_acl = ceph_get_acl,
 102         .set_acl = ceph_set_acl,
 103 };
 104
 105
 106 /*
 107  * We use a 'frag tree' to keep track of the MDS's directory fragments
 108  * for a given inode (usually there is just a single fragment).  We
 109  * need to know when a child frag is delegated to a new MDS, or when
 110  * it is flagged as replicated, so we can direct our requests
 111  * accordingly.
 112  */
 113
 114 /*
 115  * find/create a frag in the tree
 116  */
 117 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 118                                                     u32 f)
 119 {
 120         struct rb_node **p;
 121         struct rb_node *parent = NULL;
 122         struct ceph_inode_frag *frag;
 123         int c;
 124
 125         p = &ci->i_fragtree.rb_node;
 126         while (*p) {
 127                 parent = *p;
 128                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 129                 c = ceph_frag_compare(f, frag->frag);
 130                 if (c < 0)
 131                         p = &(*p)->rb_left;
 132                 else if (c > 0)
 133                         p = &(*p)->rb_right;
 134                 else
 135                         return frag;
 136         }
 137
 138         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 139         if (!frag) {
 140                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 141                        "frag %x\n", &ci->vfs_inode,
 142                        ceph_vinop(&ci->vfs_inode), f);
 143                 return ERR_PTR(-ENOMEM);
 144         }
 145         frag->frag = f;
 146         frag->split_by = 0;
 147         frag->mds = -1;
 148         frag->ndist = 0;
 149
 150         rb_link_node(&frag->node, parent, p);
 151         rb_insert_color(&frag->node, &ci->i_fragtree);
 152
 153         dout("get_or_create_frag added %llx.%llx frag %x\n",
 154              ceph_vinop(&ci->vfs_inode), f);
 155         return frag;
 156 }
 157
 158 /*
 159  * find a specific frag @f
 160  */
 161 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 162 {
 163         struct rb_node *n = ci->i_fragtree.rb_node;
 164
 165         while (n) {
 166                 struct ceph_inode_frag *frag =
 167                         rb_entry(n, struct ceph_inode_frag, node);
 168                 int c = ceph_frag_compare(f, frag->frag);
 169                 if (c < 0)
 170                         n = n->rb_left;
 171                 else if (c > 0)
 172                         n = n->rb_right;
 173                 else
 174                         return frag;
 175         }
 176         return NULL;
 177 }
 178
 179 /*
 180  * Choose frag containing the given value @v.  If @pfrag is
 181  * specified, copy the frag delegation info to the caller if
 182  * it is present.
 183  */
 184 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 185                               struct ceph_inode_frag *pfrag, int *found)
 186 {
 187         u32 t = ceph_frag_make(0, 0);
 188         struct ceph_inode_frag *frag;
 189         unsigned nway, i;
 190         u32 n;
 191
 192         if (found)
 193                 *found = 0;
 194
 195         while (1) {
 196                 WARN_ON(!ceph_frag_contains_value(t, v));
 197                 frag = __ceph_find_frag(ci, t);
 198                 if (!frag)
 199                         break; /* t is a leaf */
 200                 if (frag->split_by == 0) {
 201                         if (pfrag)
 202                                 memcpy(pfrag, frag, sizeof(*pfrag));
 203                         if (found)
 204                                 *found = 1;
 205                         break;
 206                 }
 207
 208                 /* choose child */
 209                 nway = 1 << frag->split_by;
 210                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 211                      frag->split_by, nway);
 212                 for (i = 0; i < nway; i++) {
 213                         n = ceph_frag_make_child(t, frag->split_by, i);
 214                         if (ceph_frag_contains_value(n, v)) {
 215                                 t = n;
 216                                 break;
 217                         }
 218                 }
 219                 BUG_ON(i == nway);
 220         }
 221         dout("choose_frag(%x) = %x\n", v, t);
 222
 223         return t;
 224 }
 225
 226 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 227                      struct ceph_inode_frag *pfrag, int *found)
 228 {
 229         u32 ret;
 230         mutex_lock(&ci->i_fragtree_mutex);
 231         ret = __ceph_choose_frag(ci, v, pfrag, found);
 232         mutex_unlock(&ci->i_fragtree_mutex);
 233         return ret;
 234 }
 235
 236 /*
 237  * Process dirfrag (delegation) info from the mds.  Include leaf
 238  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 239  * branches/splits are included in i_fragtree)
 240  */
 241 static int ceph_fill_dirfrag(struct inode *inode,
 242                              struct ceph_mds_reply_dirfrag *dirinfo)
 243 {
 244         struct ceph_inode_info *ci = ceph_inode(inode);
 245         struct ceph_inode_frag *frag;
 246         u32 id = le32_to_cpu(dirinfo->frag);
 247         int mds = le32_to_cpu(dirinfo->auth);
 248         int ndist = le32_to_cpu(dirinfo->ndist);
 249         int diri_auth = -1;
 250         int i;
 251         int err = 0;
 252
 253         spin_lock(&ci->i_ceph_lock);
 254         if (ci->i_auth_cap)
 255                 diri_auth = ci->i_auth_cap->mds;
 256         spin_unlock(&ci->i_ceph_lock);
 257
 258         if (mds == -1) /* CDIR_AUTH_PARENT */
 259                 mds = diri_auth;
 260
 261         mutex_lock(&ci->i_fragtree_mutex);
 262         if (ndist == 0 && mds == diri_auth) {
 263                 /* no delegation info needed. */
 264                 frag = __ceph_find_frag(ci, id);
 265                 if (!frag)
 266                         goto out;
 267                 if (frag->split_by == 0) {
 268                         /* tree leaf, remove */
 269                         dout("fill_dirfrag removed %llx.%llx frag %x"
 270                              " (no ref)\n", ceph_vinop(inode), id);
 271                         rb_erase(&frag->node, &ci->i_fragtree);
 272                         kfree(frag);
 273                 } else {
 274                         /* tree branch, keep and clear */
 275                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 276                              " referral\n", ceph_vinop(inode), id);
 277                         frag->mds = -1;
 278                         frag->ndist = 0;
 279                 }
 280                 goto out;
 281         }
 282
 283
 284         /* find/add this frag to store mds delegation info */
 285         frag = __get_or_create_frag(ci, id);
 286         if (IS_ERR(frag)) {
 287                 /* this is not the end of the world; we can continue
 288                    with bad/inaccurate delegation info */
 289                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 290                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 291                 err = -ENOMEM;
 292                 goto out;
 293         }
 294
 295         frag->mds = mds;
 296         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 297         for (i = 0; i < frag->ndist; i++)
 298                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 299         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 300              ceph_vinop(inode), frag->frag, frag->ndist);
 301
 302 out:
 303         mutex_unlock(&ci->i_fragtree_mutex);
 304         return err;
 305 }
 306
 307 static int frag_tree_split_cmp(const void *l, const void *r)
 308 {
 309         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 310         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 311         return ceph_frag_compare(ls->frag, rs->frag);
 312 }
 313
 314 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 315 {
 316         if (!frag)
 317                 return f == ceph_frag_make(0, 0);
 318         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 319                 return false;
 320         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 321 }
 322
 323 static int ceph_fill_fragtree(struct inode *inode,
 324                               struct ceph_frag_tree_head *fragtree,
 325                               struct ceph_mds_reply_dirfrag *dirinfo)
 326 {
 327         struct ceph_inode_info *ci = ceph_inode(inode);
 328         struct ceph_inode_frag *frag, *prev_frag = NULL;
 329         struct rb_node *rb_node;
 330         unsigned i, split_by, nsplits;
 331         u32 id;
 332         bool update = false;
 333
 334         mutex_lock(&ci->i_fragtree_mutex);
 335         nsplits = le32_to_cpu(fragtree->nsplits);
 336         if (nsplits != ci->i_fragtree_nsplits) {
 337                 update = true;
 338         } else if (nsplits) {
 339                 i = prandom_u32() % nsplits;
 340                 id = le32_to_cpu(fragtree->splits[i].frag);
 341                 if (!__ceph_find_frag(ci, id))
 342                         update = true;
 343         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 344                 rb_node = rb_first(&ci->i_fragtree);
 345                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 346                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 347                         update = true;
 348         }
 349         if (!update && dirinfo) {
 350                 id = le32_to_cpu(dirinfo->frag);
 351                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 352                         update = true;
 353         }
 354         if (!update)
 355                 goto out_unlock;
 356
 357         if (nsplits > 1) {
 358                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 359                      frag_tree_split_cmp, NULL);
 360         }
 361
 362         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 363         rb_node = rb_first(&ci->i_fragtree);
 364         for (i = 0; i < nsplits; i++) {
 365                 id = le32_to_cpu(fragtree->splits[i].frag);
 366                 split_by = le32_to_cpu(fragtree->splits[i].by);
 367                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 368                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 369                                "frag %x split by %d\n", ceph_vinop(inode),
 370                                i, nsplits, id, split_by);
 371                         continue;
 372                 }
 373                 frag = NULL;
 374                 while (rb_node) {
 375                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 376                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 377                                 if (frag->frag != id)
 378                                         frag = NULL;
 379                                 else
 380                                         rb_node = rb_next(rb_node);
 381                                 break;
 382                         }
 383                         rb_node = rb_next(rb_node);
 384                         /* delete stale split/leaf node */
 385                         if (frag->split_by > 0 ||
 386                             !is_frag_child(frag->frag, prev_frag)) {
 387                                 rb_erase(&frag->node, &ci->i_fragtree);
 388                                 if (frag->split_by > 0)
 389                                         ci->i_fragtree_nsplits--;
 390                                 kfree(frag);
 391                         }
 392                         frag = NULL;
 393                 }
 394                 if (!frag) {
 395                         frag = __get_or_create_frag(ci, id);
 396                         if (IS_ERR(frag))
 397                                 continue;
 398                 }
 399                 if (frag->split_by == 0)
 400                         ci->i_fragtree_nsplits++;
 401                 frag->split_by = split_by;
 402                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 403                 prev_frag = frag;
 404         }
 405         while (rb_node) {
 406                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 407                 rb_node = rb_next(rb_node);
 408                 /* delete stale split/leaf node */
 409                 if (frag->split_by > 0 ||
 410                     !is_frag_child(frag->frag, prev_frag)) {
 411                         rb_erase(&frag->node, &ci->i_fragtree);
 412                         if (frag->split_by > 0)
 413                                 ci->i_fragtree_nsplits--;
 414                         kfree(frag);
 415                 }
 416         }
 417 out_unlock:
 418         mutex_unlock(&ci->i_fragtree_mutex);
 419         return 0;
 420 }
 421
 422 /*
 423  * initialize a newly allocated inode.
 424  */
 425 struct inode *ceph_alloc_inode(struct super_block *sb)
 426 {
 427         struct ceph_inode_info *ci;
 428         int i;
 429
 430         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 431         if (!ci)
 432                 return NULL;
 433
 434         dout("alloc_inode %p\n", &ci->vfs_inode);
 435
 436         spin_lock_init(&ci->i_ceph_lock);
 437
 438         ci->i_version = 0;
 439         ci->i_inline_version = 0;
 440         ci->i_time_warp_seq = 0;
 441         ci->i_ceph_flags = 0;
 442         atomic64_set(&ci->i_ordered_count, 1);
 443         atomic64_set(&ci->i_release_count, 1);
 444         atomic64_set(&ci->i_complete_seq[0], 0);
 445         atomic64_set(&ci->i_complete_seq[1], 0);
 446         ci->i_symlink = NULL;
 447
 448         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 449         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 450
 451         ci->i_fragtree = RB_ROOT;
 452         mutex_init(&ci->i_fragtree_mutex);
 453
 454         ci->i_xattrs.blob = NULL;
 455         ci->i_xattrs.prealloc_blob = NULL;
 456         ci->i_xattrs.dirty = false;
 457         ci->i_xattrs.index = RB_ROOT;
 458         ci->i_xattrs.count = 0;
 459         ci->i_xattrs.names_size = 0;
 460         ci->i_xattrs.vals_size = 0;
 461         ci->i_xattrs.version = 0;
 462         ci->i_xattrs.index_version = 0;
 463
 464         ci->i_caps = RB_ROOT;
 465         ci->i_auth_cap = NULL;
 466         ci->i_dirty_caps = 0;
 467         ci->i_flushing_caps = 0;
 468         INIT_LIST_HEAD(&ci->i_dirty_item);
 469         INIT_LIST_HEAD(&ci->i_flushing_item);
 470         ci->i_prealloc_cap_flush = NULL;
 471         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 472         init_waitqueue_head(&ci->i_cap_wq);
 473         ci->i_hold_caps_min = 0;
 474         ci->i_hold_caps_max = 0;
 475         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 476         INIT_LIST_HEAD(&ci->i_cap_snaps);
 477         ci->i_head_snapc = NULL;
 478         ci->i_snap_caps = 0;
 479
 480         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 481                 ci->i_nr_by_mode[i] = 0;
 482
 483         mutex_init(&ci->i_truncate_mutex);
 484         ci->i_truncate_seq = 0;
 485         ci->i_truncate_size = 0;
 486         ci->i_truncate_pending = 0;
 487
 488         ci->i_max_size = 0;
 489         ci->i_reported_size = 0;
 490         ci->i_wanted_max_size = 0;
 491         ci->i_requested_max_size = 0;
 492
 493         ci->i_pin_ref = 0;
 494         ci->i_rd_ref = 0;
 495         ci->i_rdcache_ref = 0;
 496         ci->i_wr_ref = 0;
 497         ci->i_wb_ref = 0;
 498         ci->i_wrbuffer_ref = 0;
 499         ci->i_wrbuffer_ref_head = 0;
 500         ci->i_shared_gen = 0;
 501         ci->i_rdcache_gen = 0;
 502         ci->i_rdcache_revoking = 0;
 503
 504         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 505         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 506         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 507         spin_lock_init(&ci->i_unsafe_lock);
 508
 509         ci->i_snap_realm = NULL;
 510         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 511         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 512
 513         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 514         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 515
 516         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 517
 518         ceph_fscache_inode_init(ci);
 519
 520         return &ci->vfs_inode;
 521 }
 522
 523 static void ceph_i_callback(struct rcu_head *head)
 524 {
 525         struct inode *inode = container_of(head, struct inode, i_rcu);
 526         struct ceph_inode_info *ci = ceph_inode(inode);
 527
 528         kmem_cache_free(ceph_inode_cachep, ci);
 529 }
 530
 531 void ceph_destroy_inode(struct inode *inode)
 532 {
 533         struct ceph_inode_info *ci = ceph_inode(inode);
 534         struct ceph_inode_frag *frag;
 535         struct rb_node *n;
 536
 537         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 538
 539         ceph_fscache_unregister_inode_cookie(ci);
 540
 541         ceph_queue_caps_release(inode);
 542
 543         /*
 544          * we may still have a snap_realm reference if there are stray
 545          * caps in i_snap_caps.
 546          */
 547         if (ci->i_snap_realm) {
 548                 struct ceph_mds_client *mdsc =
 549                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 550                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 551
 552                 dout(" dropping residual ref to snap realm %p\n", realm);
 553                 spin_lock(&realm->inodes_with_caps_lock);
 554                 list_del_init(&ci->i_snap_realm_item);
 555                 spin_unlock(&realm->inodes_with_caps_lock);
 556                 ceph_put_snap_realm(mdsc, realm);
 557         }
 558
 559         kfree(ci->i_symlink);
 560         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 561                 frag = rb_entry(n, struct ceph_inode_frag, node);
 562                 rb_erase(n, &ci->i_fragtree);
 563                 kfree(frag);
 564         }
 565         ci->i_fragtree_nsplits = 0;
 566
 567         __ceph_destroy_xattrs(ci);
 568         if (ci->i_xattrs.blob)
 569                 ceph_buffer_put(ci->i_xattrs.blob);
 570         if (ci->i_xattrs.prealloc_blob)
 571                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 572
 573         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 574
 575         call_rcu(&inode->i_rcu, ceph_i_callback);
 576 }
 577
 578 int ceph_drop_inode(struct inode *inode)
 579 {
 580         /*
 581          * Positve dentry and corresponding inode are always accompanied
 582          * in MDS reply. So no need to keep inode in the cache after
 583          * dropping all its aliases.
 584          */
 585         return 1;
 586 }
 587
 588 void ceph_evict_inode(struct inode *inode)
 589 {
 590         /* wait unsafe sync writes */
 591         ceph_sync_write_wait(inode);
 592         truncate_inode_pages_final(&inode->i_data);
 593         clear_inode(inode);
 594 }
 595
 596 static inline blkcnt_t calc_inode_blocks(u64 size)
 597 {
 598         return (size + (1<<9) - 1) >> 9;
 599 }
 600
 601 /*
 602  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 603  * careful because either the client or MDS may have more up to date
 604  * info, depending on which capabilities are held, and whether
 605  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 606  * and size are monotonically increasing, except when utimes() or
 607  * truncate() increments the corresponding _seq values.)
 608  */
 609 int ceph_fill_file_size(struct inode *inode, int issued,
 610                         u32 truncate_seq, u64 truncate_size, u64 size)
 611 {
 612         struct ceph_inode_info *ci = ceph_inode(inode);
 613         int queue_trunc = 0;
 614
 615         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 616             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 617                 dout("size %lld -> %llu\n", inode->i_size, size);
 618                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 619                         pr_err("fill_file_size non-zero size for directory\n");
 620                         size = 0;
 621                 }
 622                 i_size_write(inode, size);
 623                 inode->i_blocks = calc_inode_blocks(size);
 624                 ci->i_reported_size = size;
 625                 if (truncate_seq != ci->i_truncate_seq) {
 626                         dout("truncate_seq %u -> %u\n",
 627                              ci->i_truncate_seq, truncate_seq);
 628                         ci->i_truncate_seq = truncate_seq;
 629
 630                         /* the MDS should have revoked these caps */
 631                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 632                                                CEPH_CAP_FILE_RD |
 633                                                CEPH_CAP_FILE_WR |
 634                                                CEPH_CAP_FILE_LAZYIO));
 635                         /*
 636                          * If we hold relevant caps, or in the case where we're
 637                          * not the only client referencing this file and we
 638                          * don't hold those caps, then we need to check whether
 639                          * the file is either opened or mmaped
 640                          */
 641                         if ((issued & (CEPH_CAP_FILE_CACHE|
 642                                        CEPH_CAP_FILE_BUFFER)) ||
 643                             mapping_mapped(inode->i_mapping) ||
 644                             __ceph_caps_file_wanted(ci)) {
 645                                 ci->i_truncate_pending++;
 646                                 queue_trunc = 1;
 647                         }
 648                 }
 649         }
 650         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 651             ci->i_truncate_size != truncate_size) {
 652                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 653                      truncate_size);
 654                 ci->i_truncate_size = truncate_size;
 655         }
 656
 657         if (queue_trunc)
 658                 ceph_fscache_invalidate(inode);
 659
 660         return queue_trunc;
 661 }
 662
 663 void ceph_fill_file_time(struct inode *inode, int issued,
 664                          u64 time_warp_seq, struct timespec *ctime,
 665                          struct timespec *mtime, struct timespec *atime)
 666 {
 667         struct ceph_inode_info *ci = ceph_inode(inode);
 668         int warn = 0;
 669
 670         if (issued & (CEPH_CAP_FILE_EXCL|
 671                       CEPH_CAP_FILE_WR|
 672                       CEPH_CAP_FILE_BUFFER|
 673                       CEPH_CAP_AUTH_EXCL|
 674                       CEPH_CAP_XATTR_EXCL)) {
 675                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 676                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 677                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 678                              ctime->tv_sec, ctime->tv_nsec);
 679                         inode->i_ctime = *ctime;
 680                 }
 681                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 682                         /* the MDS did a utimes() */
 683                         dout("mtime %ld.%09ld -> %ld.%09ld "
 684                              "tw %d -> %d\n",
 685                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 686                              mtime->tv_sec, mtime->tv_nsec,
 687                              ci->i_time_warp_seq, (int)time_warp_seq);
 688
 689                         inode->i_mtime = *mtime;
 690                         inode->i_atime = *atime;
 691                         ci->i_time_warp_seq = time_warp_seq;
 692                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 693                         /* nobody did utimes(); take the max */
 694                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 695                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 696                                      inode->i_mtime.tv_sec,
 697                                      inode->i_mtime.tv_nsec,
 698                                      mtime->tv_sec, mtime->tv_nsec);
 699                                 inode->i_mtime = *mtime;
 700                         }
 701                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 702                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 703                                      inode->i_atime.tv_sec,
 704                                      inode->i_atime.tv_nsec,
 705                                      atime->tv_sec, atime->tv_nsec);
 706                                 inode->i_atime = *atime;
 707                         }
 708                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 709                         /* we did a utimes(); ignore mds values */
 710                 } else {
 711                         warn = 1;
 712                 }
 713         } else {
 714                 /* we have no write|excl caps; whatever the MDS says is true */
 715                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 716                         inode->i_ctime = *ctime;
 717                         inode->i_mtime = *mtime;
 718                         inode->i_atime = *atime;
 719                         ci->i_time_warp_seq = time_warp_seq;
 720                 } else {
 721                         warn = 1;
 722                 }
 723         }
 724         if (warn) /* time_warp_seq shouldn't go backwards */
 725                 dout("%p mds time_warp_seq %llu < %u\n",
 726                      inode, time_warp_seq, ci->i_time_warp_seq);
 727 }
 728
 729 /*
 730  * Populate an inode based on info from mds.  May be called on new or
 731  * existing inodes.
 732  */
 733 static int fill_inode(struct inode *inode, struct page *locked_page,
 734                       struct ceph_mds_reply_info_in *iinfo,
 735                       struct ceph_mds_reply_dirfrag *dirinfo,
 736                       struct ceph_mds_session *session,
 737                       unsigned long ttl_from, int cap_fmode,
 738                       struct ceph_cap_reservation *caps_reservation)
 739 {
 740         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 741         struct ceph_mds_reply_inode *info = iinfo->in;
 742         struct ceph_inode_info *ci = ceph_inode(inode);
 743         int issued = 0, implemented, new_issued;
 744         struct timespec mtime, atime, ctime;
 745         struct ceph_buffer *xattr_blob = NULL;
 746         struct ceph_string *pool_ns = NULL;
 747         struct ceph_cap *new_cap = NULL;
 748         int err = 0;
 749         bool wake = false;
 750         bool queue_trunc = false;
 751         bool new_version = false;
 752         bool fill_inline = false;
 753
 754         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 755              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 756              ci->i_version);
 757
 758         /* prealloc new cap struct */
 759         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 760                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 761
 762         /*
 763          * prealloc xattr data, if it looks like we'll need it.  only
 764          * if len > 4 (meaning there are actually xattrs; the first 4
 765          * bytes are the xattr count).
 766          */
 767         if (iinfo->xattr_len > 4) {
 768                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 769                 if (!xattr_blob)
 770                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 771                                iinfo->xattr_len);
 772         }
 773
 774         if (iinfo->pool_ns_len > 0)
 775                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 776                                                      iinfo->pool_ns_len);
 777
 778         spin_lock(&ci->i_ceph_lock);
 779
 780         /*
 781          * provided version will be odd if inode value is projected,
 782          * even if stable.  skip the update if we have newer stable
 783          * info (ours>=theirs, e.g. due to racing mds replies), unless
 784          * we are getting projected (unstable) info (in which case the
 785          * version is odd, and we want ours>theirs).
 786          *   us   them
 787          *   2    2     skip
 788          *   3    2     skip
 789          *   3    3     update
 790          */
 791         if (ci->i_version == 0 ||
 792             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 793              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 794                 new_version = true;
 795
 796         issued = __ceph_caps_issued(ci, &implemented);
 797         issued |= implemented | __ceph_caps_dirty(ci);
 798         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 799
 800         /* update inode */
 801         ci->i_version = le64_to_cpu(info->version);
 802         inode->i_version++;
 803         inode->i_rdev = le32_to_cpu(info->rdev);
 804         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 805
 806         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 807             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 808                 inode->i_mode = le32_to_cpu(info->mode);
 809                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 810                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 811                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 812                      from_kuid(&init_user_ns, inode->i_uid),
 813                      from_kgid(&init_user_ns, inode->i_gid));
 814         }
 815
 816         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 817             (issued & CEPH_CAP_LINK_EXCL) == 0)
 818                 set_nlink(inode, le32_to_cpu(info->nlink));
 819
 820         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 821                 /* be careful with mtime, atime, size */
 822                 ceph_decode_timespec(&atime, &info->atime);
 823                 ceph_decode_timespec(&mtime, &info->mtime);
 824                 ceph_decode_timespec(&ctime, &info->ctime);
 825                 ceph_fill_file_time(inode, issued,
 826                                 le32_to_cpu(info->time_warp_seq),
 827                                 &ctime, &mtime, &atime);
 828         }
 829
 830         if (new_version ||
 831             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 832                 s64 old_pool = ci->i_layout.pool_id;
 833                 struct ceph_string *old_ns;
 834
 835                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 836                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 837                                         lockdep_is_held(&ci->i_ceph_lock));
 838                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 839
 840                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 841                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 842
 843                 pool_ns = old_ns;
 844
 845                 queue_trunc = ceph_fill_file_size(inode, issued,
 846                                         le32_to_cpu(info->truncate_seq),
 847                                         le64_to_cpu(info->truncate_size),
 848                                         le64_to_cpu(info->size));
 849                 /* only update max_size on auth cap */
 850                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 851                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 852                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 853                                         le64_to_cpu(info->max_size));
 854                         ci->i_max_size = le64_to_cpu(info->max_size);
 855                 }
 856         }
 857
 858         /* xattrs */
 859         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 860         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 861             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 862                 if (ci->i_xattrs.blob)
 863                         ceph_buffer_put(ci->i_xattrs.blob);
 864                 ci->i_xattrs.blob = xattr_blob;
 865                 if (xattr_blob)
 866                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 867                                iinfo->xattr_data, iinfo->xattr_len);
 868                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 869                 ceph_forget_all_cached_acls(inode);
 870                 xattr_blob = NULL;
 871         }
 872
 873         inode->i_mapping->a_ops = &ceph_aops;
 874
 875         switch (inode->i_mode & S_IFMT) {
 876         case S_IFIFO:
 877         case S_IFBLK:
 878         case S_IFCHR:
 879         case S_IFSOCK:
 880                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 881                 inode->i_op = &ceph_file_iops;
 882                 break;
 883         case S_IFREG:
 884                 inode->i_op = &ceph_file_iops;
 885                 inode->i_fop = &ceph_file_fops;
 886                 break;
 887         case S_IFLNK:
 888                 inode->i_op = &ceph_symlink_iops;
 889                 if (!ci->i_symlink) {
 890                         u32 symlen = iinfo->symlink_len;
 891                         char *sym;
 892
 893                         spin_unlock(&ci->i_ceph_lock);
 894
 895                         if (symlen != i_size_read(inode)) {
 896                                 pr_err("fill_inode %llx.%llx BAD symlink "
 897                                         "size %lld\n", ceph_vinop(inode),
 898                                         i_size_read(inode));
 899                                 i_size_write(inode, symlen);
 900                                 inode->i_blocks = calc_inode_blocks(symlen);
 901                         }
 902
 903                         err = -ENOMEM;
 904                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 905                         if (!sym)
 906                                 goto out;
 907
 908                         spin_lock(&ci->i_ceph_lock);
 909                         if (!ci->i_symlink)
 910                                 ci->i_symlink = sym;
 911                         else
 912                                 kfree(sym); /* lost a race */
 913                 }
 914                 inode->i_link = ci->i_symlink;
 915                 break;
 916         case S_IFDIR:
 917                 inode->i_op = &ceph_dir_iops;
 918                 inode->i_fop = &ceph_dir_fops;
 919
 920                 ci->i_dir_layout = iinfo->dir_layout;
 921
 922                 ci->i_files = le64_to_cpu(info->files);
 923                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 924                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 925                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 926                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 927                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 928                 break;
 929         default:
 930                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 931                        ceph_vinop(inode), inode->i_mode);
 932         }
 933
 934         /* were we issued a capability? */
 935         if (info->cap.caps) {
 936                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 937                         unsigned caps = le32_to_cpu(info->cap.caps);
 938                         ceph_add_cap(inode, session,
 939                                      le64_to_cpu(info->cap.cap_id),
 940                                      cap_fmode, caps,
 941                                      le32_to_cpu(info->cap.wanted),
 942                                      le32_to_cpu(info->cap.seq),
 943                                      le32_to_cpu(info->cap.mseq),
 944                                      le64_to_cpu(info->cap.realm),
 945                                      info->cap.flags, &new_cap);
 946
 947                         /* set dir completion flag? */
 948                         if (S_ISDIR(inode->i_mode) &&
 949                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 950                             (caps & CEPH_CAP_FILE_SHARED) &&
 951                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 952                             !__ceph_dir_is_complete(ci)) {
 953                                 dout(" marking %p complete (empty)\n", inode);
 954                                 i_size_write(inode, 0);
 955                                 __ceph_dir_set_complete(ci,
 956                                         atomic64_read(&ci->i_release_count),
 957                                         atomic64_read(&ci->i_ordered_count));
 958                         }
 959
 960                         wake = true;
 961                 } else {
 962                         dout(" %p got snap_caps %s\n", inode,
 963                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 964                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 965                         if (cap_fmode >= 0)
 966                                 __ceph_get_fmode(ci, cap_fmode);
 967                 }
 968         } else if (cap_fmode >= 0) {
 969                 pr_warn("mds issued no caps on %llx.%llx\n",
 970                            ceph_vinop(inode));
 971                 __ceph_get_fmode(ci, cap_fmode);
 972         }
 973
 974         if (iinfo->inline_version > 0 &&
 975             iinfo->inline_version >= ci->i_inline_version) {
 976                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 977                 ci->i_inline_version = iinfo->inline_version;
 978                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 979                     (locked_page ||
 980                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 981                         fill_inline = true;
 982         }
 983
 984         spin_unlock(&ci->i_ceph_lock);
 985
 986         if (fill_inline)
 987                 ceph_fill_inline_data(inode, locked_page,
 988                                       iinfo->inline_data, iinfo->inline_len);
 989
 990         if (wake)
 991                 wake_up_all(&ci->i_cap_wq);
 992
 993         /* queue truncate if we saw i_size decrease */
 994         if (queue_trunc)
 995                 ceph_queue_vmtruncate(inode);
 996
 997         /* populate frag tree */
 998         if (S_ISDIR(inode->i_mode))
 999                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1000
1001         /* update delegation info? */
1002         if (dirinfo)
1003                 ceph_fill_dirfrag(inode, dirinfo);
1004
1005         err = 0;
1006 out:
1007         if (new_cap)
1008                 ceph_put_cap(mdsc, new_cap);
1009         if (xattr_blob)
1010                 ceph_buffer_put(xattr_blob);
1011         ceph_put_string(pool_ns);
1012         return err;
1013 }
1014
1015 /*
1016  * caller should hold session s_mutex.
1017  */
1018 static void update_dentry_lease(struct dentry *dentry,
1019                                 struct ceph_mds_reply_lease *lease,
1020                                 struct ceph_mds_session *session,
1021                                 unsigned long from_time)
1022 {
1023         struct ceph_dentry_info *di = ceph_dentry(dentry);
1024         long unsigned duration = le32_to_cpu(lease->duration_ms);
1025         long unsigned ttl = from_time + (duration * HZ) / 1000;
1026         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1027         struct inode *dir;
1028
1029         /* only track leases on regular dentries */
1030         if (dentry->d_op != &ceph_dentry_ops)
1031                 return;
1032
1033         spin_lock(&dentry->d_lock);
1034         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1035              dentry, duration, ttl);
1036
1037         /* make lease_rdcache_gen match directory */
1038         dir = d_inode(dentry->d_parent);
1039         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1040
1041         if (duration == 0)
1042                 goto out_unlock;
1043
1044         if (di->lease_gen == session->s_cap_gen &&
1045             time_before(ttl, di->time))
1046                 goto out_unlock;  /* we already have a newer lease. */
1047
1048         if (di->lease_session && di->lease_session != session)
1049                 goto out_unlock;
1050
1051         ceph_dentry_lru_touch(dentry);
1052
1053         if (!di->lease_session)
1054                 di->lease_session = ceph_get_mds_session(session);
1055         di->lease_gen = session->s_cap_gen;
1056         di->lease_seq = le32_to_cpu(lease->seq);
1057         di->lease_renew_after = half_ttl;
1058         di->lease_renew_from = 0;
1059         di->time = ttl;
1060 out_unlock:
1061         spin_unlock(&dentry->d_lock);
1062         return;
1063 }
1064
1065 /*
1066  * splice a dentry to an inode.
1067  * caller must hold directory i_mutex for this to be safe.
1068  */
1069 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1070 {
1071         struct dentry *realdn;
1072
1073         BUG_ON(d_inode(dn));
1074
1075         /* dn must be unhashed */
1076         if (!d_unhashed(dn))
1077                 d_drop(dn);
1078         realdn = d_splice_alias(in, dn);
1079         if (IS_ERR(realdn)) {
1080                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1081                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1082                 dn = realdn; /* note realdn contains the error */
1083                 goto out;
1084         } else if (realdn) {
1085                 dout("dn %p (%d) spliced with %p (%d) "
1086                      "inode %p ino %llx.%llx\n",
1087                      dn, d_count(dn),
1088                      realdn, d_count(realdn),
1089                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1090                 dput(dn);
1091                 dn = realdn;
1092         } else {
1093                 BUG_ON(!ceph_dentry(dn));
1094                 dout("dn %p attached to %p ino %llx.%llx\n",
1095                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1096         }
1097 out:
1098         return dn;
1099 }
1100
1101 /*
1102  * Incorporate results into the local cache.  This is either just
1103  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1104  * after a lookup).
1105  *
1106  * A reply may contain
1107  *         a directory inode along with a dentry.
1108  *  and/or a target inode
1109  *
1110  * Called with snap_rwsem (read).
1111  */
1112 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1113                     struct ceph_mds_session *session)
1114 {
1115         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1116         struct inode *in = NULL;
1117         struct ceph_vino vino;
1118         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1119         int err = 0;
1120
1121         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1122              rinfo->head->is_dentry, rinfo->head->is_target);
1123
1124 #if 0
1125         /*
1126          * Debugging hook:
1127          *
1128          * If we resend completed ops to a recovering mds, we get no
1129          * trace.  Since that is very rare, pretend this is the case
1130          * to ensure the 'no trace' handlers in the callers behave.
1131          *
1132          * Fill in inodes unconditionally to avoid breaking cap
1133          * invariants.
1134          */
1135         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1136                 pr_info("fill_trace faking empty trace on %lld %s\n",
1137                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1138                 if (rinfo->head->is_dentry) {
1139                         rinfo->head->is_dentry = 0;
1140                         err = fill_inode(req->r_locked_dir,
1141                                          &rinfo->diri, rinfo->dirfrag,
1142                                          session, req->r_request_started, -1);
1143                 }
1144                 if (rinfo->head->is_target) {
1145                         rinfo->head->is_target = 0;
1146                         ininfo = rinfo->targeti.in;
1147                         vino.ino = le64_to_cpu(ininfo->ino);
1148                         vino.snap = le64_to_cpu(ininfo->snapid);
1149                         in = ceph_get_inode(sb, vino);
1150                         err = fill_inode(in, &rinfo->targeti, NULL,
1151                                          session, req->r_request_started,
1152                                          req->r_fmode);
1153                         iput(in);
1154                 }
1155         }
1156 #endif
1157
1158         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1159                 dout("fill_trace reply is empty!\n");
1160                 if (rinfo->head->result == 0 && req->r_locked_dir)
1161                         ceph_invalidate_dir_request(req);
1162                 return 0;
1163         }
1164
1165         if (rinfo->head->is_dentry) {
1166                 struct inode *dir = req->r_locked_dir;
1167
1168                 if (dir) {
1169                         err = fill_inode(dir, NULL,
1170                                          &rinfo->diri, rinfo->dirfrag,
1171                                          session, req->r_request_started, -1,
1172                                          &req->r_caps_reservation);
1173                         if (err < 0)
1174                                 goto done;
1175                 } else {
1176                         WARN_ON_ONCE(1);
1177                 }
1178
1179                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1180                         struct qstr dname;
1181                         struct dentry *dn, *parent;
1182
1183                         BUG_ON(!rinfo->head->is_target);
1184                         BUG_ON(req->r_dentry);
1185
1186                         parent = d_find_any_alias(dir);
1187                         BUG_ON(!parent);
1188
1189                         dname.name = rinfo->dname;
1190                         dname.len = rinfo->dname_len;
1191                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1192                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1193                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1194 retry_lookup:
1195                         dn = d_lookup(parent, &dname);
1196                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1197                              parent, dname.len, dname.name, dn);
1198
1199                         if (!dn) {
1200                                 dn = d_alloc(parent, &dname);
1201                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1202                                      dname.len, dname.name, dn);
1203                                 if (dn == NULL) {
1204                                         dput(parent);
1205                                         err = -ENOMEM;
1206                                         goto done;
1207                                 }
1208                                 err = ceph_init_dentry(dn);
1209                                 if (err < 0) {
1210                                         dput(dn);
1211                                         dput(parent);
1212                                         goto done;
1213                                 }
1214                         } else if (d_really_is_positive(dn) &&
1215                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1216                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1217                                 dout(" dn %p points to wrong inode %p\n",
1218                                      dn, d_inode(dn));
1219                                 d_delete(dn);
1220                                 dput(dn);
1221                                 goto retry_lookup;
1222                         }
1223
1224                         req->r_dentry = dn;
1225                         dput(parent);
1226                 }
1227         }
1228
1229         if (rinfo->head->is_target) {
1230                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1231                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1232
1233                 in = ceph_get_inode(sb, vino);
1234                 if (IS_ERR(in)) {
1235                         err = PTR_ERR(in);
1236                         goto done;
1237                 }
1238                 req->r_target_inode = in;
1239
1240                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1241                                 session, req->r_request_started,
1242                                 (!req->r_aborted && rinfo->head->result == 0) ?
1243                                 req->r_fmode : -1,
1244                                 &req->r_caps_reservation);
1245                 if (err < 0) {
1246                         pr_err("fill_inode badness %p %llx.%llx\n",
1247                                 in, ceph_vinop(in));
1248                         goto done;
1249                 }
1250         }
1251
1252         /*
1253          * ignore null lease/binding on snapdir ENOENT, or else we
1254          * will have trouble splicing in the virtual snapdir later
1255          */
1256         if (rinfo->head->is_dentry && !req->r_aborted &&
1257             req->r_locked_dir &&
1258             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1259                                                fsc->mount_options->snapdir_name,
1260                                                req->r_dentry->d_name.len))) {
1261                 /*
1262                  * lookup link rename   : null -> possibly existing inode
1263                  * mknod symlink mkdir  : null -> new inode
1264                  * unlink               : linked -> null
1265                  */
1266                 struct inode *dir = req->r_locked_dir;
1267                 struct dentry *dn = req->r_dentry;
1268                 bool have_dir_cap, have_lease;
1269
1270                 BUG_ON(!dn);
1271                 BUG_ON(!dir);
1272                 BUG_ON(d_inode(dn->d_parent) != dir);
1273                 BUG_ON(ceph_ino(dir) !=
1274                        le64_to_cpu(rinfo->diri.in->ino));
1275                 BUG_ON(ceph_snap(dir) !=
1276                        le64_to_cpu(rinfo->diri.in->snapid));
1277
1278                 /* do we have a lease on the whole dir? */
1279                 have_dir_cap =
1280                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1281                          CEPH_CAP_FILE_SHARED);
1282
1283                 /* do we have a dn lease? */
1284                 have_lease = have_dir_cap ||
1285                         le32_to_cpu(rinfo->dlease->duration_ms);
1286                 if (!have_lease)
1287                         dout("fill_trace  no dentry lease or dir cap\n");
1288
1289                 /* rename? */
1290                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1291                         struct inode *olddir = req->r_old_dentry_dir;
1292                         BUG_ON(!olddir);
1293
1294                         dout(" src %p '%pd' dst %p '%pd'\n",
1295                              req->r_old_dentry,
1296                              req->r_old_dentry,
1297                              dn, dn);
1298                         dout("fill_trace doing d_move %p -> %p\n",
1299                              req->r_old_dentry, dn);
1300
1301                         /* d_move screws up sibling dentries' offsets */
1302                         ceph_dir_clear_ordered(dir);
1303                         ceph_dir_clear_ordered(olddir);
1304
1305                         d_move(req->r_old_dentry, dn);
1306                         dout(" src %p '%pd' dst %p '%pd'\n",
1307                              req->r_old_dentry,
1308                              req->r_old_dentry,
1309                              dn, dn);
1310
1311                         /* ensure target dentry is invalidated, despite
1312                            rehashing bug in vfs_rename_dir */
1313                         ceph_invalidate_dentry_lease(dn);
1314
1315                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1316                              ceph_dentry(req->r_old_dentry)->offset);
1317
1318                         dn = req->r_old_dentry;  /* use old_dentry */
1319                 }
1320
1321                 /* null dentry? */
1322                 if (!rinfo->head->is_target) {
1323                         dout("fill_trace null dentry\n");
1324                         if (d_really_is_positive(dn)) {
1325                                 ceph_dir_clear_ordered(dir);
1326                                 dout("d_delete %p\n", dn);
1327                                 d_delete(dn);
1328                         } else {
1329                                 if (have_lease && d_unhashed(dn))
1330                                         d_add(dn, NULL);
1331                                 update_dentry_lease(dn, rinfo->dlease,
1332                                                     session,
1333                                                     req->r_request_started);
1334                         }
1335                         goto done;
1336                 }
1337
1338                 /* attach proper inode */
1339                 if (d_really_is_negative(dn)) {
1340                         ceph_dir_clear_ordered(dir);
1341                         ihold(in);
1342                         dn = splice_dentry(dn, in);
1343                         if (IS_ERR(dn)) {
1344                                 err = PTR_ERR(dn);
1345                                 goto done;
1346                         }
1347                         req->r_dentry = dn;  /* may have spliced */
1348                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1349                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1350                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1351                              ceph_vinop(in));
1352                         d_invalidate(dn);
1353                         have_lease = false;
1354                 }
1355
1356                 if (have_lease)
1357                         update_dentry_lease(dn, rinfo->dlease, session,
1358                                             req->r_request_started);
1359                 dout(" final dn %p\n", dn);
1360         } else if (!req->r_aborted &&
1361                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1362                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1363                 struct dentry *dn = req->r_dentry;
1364                 struct inode *dir = req->r_locked_dir;
1365
1366                 /* fill out a snapdir LOOKUPSNAP dentry */
1367                 BUG_ON(!dn);
1368                 BUG_ON(!dir);
1369                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1370                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1371                 ceph_dir_clear_ordered(dir);
1372                 ihold(in);
1373                 dn = splice_dentry(dn, in);
1374                 if (IS_ERR(dn)) {
1375                         err = PTR_ERR(dn);
1376                         goto done;
1377                 }
1378                 req->r_dentry = dn;  /* may have spliced */
1379         }
1380 done:
1381         dout("fill_trace done err=%d\n", err);
1382         return err;
1383 }
1384
1385 /*
1386  * Prepopulate our cache with readdir results, leases, etc.
1387  */
1388 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1389                                            struct ceph_mds_session *session)
1390 {
1391         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1392         int i, err = 0;
1393
1394         for (i = 0; i < rinfo->dir_nr; i++) {
1395                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1396                 struct ceph_vino vino;
1397                 struct inode *in;
1398                 int rc;
1399
1400                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1401                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1402
1403                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1404                 if (IS_ERR(in)) {
1405                         err = PTR_ERR(in);
1406                         dout("new_inode badness got %d\n", err);
1407                         continue;
1408                 }
1409                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1410                                 req->r_request_started, -1,
1411                                 &req->r_caps_reservation);
1412                 if (rc < 0) {
1413                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1414                         err = rc;
1415                 }
1416                 iput(in);
1417         }
1418
1419         return err;
1420 }
1421
1422 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1423 {
1424         if (ctl->page) {
1425                 kunmap(ctl->page);
1426                 put_page(ctl->page);
1427                 ctl->page = NULL;
1428         }
1429 }
1430
1431 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1432                               struct ceph_readdir_cache_control *ctl,
1433                               struct ceph_mds_request *req)
1434 {
1435         struct ceph_inode_info *ci = ceph_inode(dir);
1436         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1437         unsigned idx = ctl->index % nsize;
1438         pgoff_t pgoff = ctl->index / nsize;
1439
1440         if (!ctl->page || pgoff != page_index(ctl->page)) {
1441                 ceph_readdir_cache_release(ctl);
1442                 if (idx == 0)
1443                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1444                 else
1445                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1446                 if (!ctl->page) {
1447                         ctl->index = -1;
1448                         return idx == 0 ? -ENOMEM : 0;
1449                 }
1450                 /* reading/filling the cache are serialized by
1451                  * i_mutex, no need to use page lock */
1452                 unlock_page(ctl->page);
1453                 ctl->dentries = kmap(ctl->page);
1454                 if (idx == 0)
1455                         memset(ctl->dentries, 0, PAGE_SIZE);
1456         }
1457
1458         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1459             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1460                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1461                 ctl->dentries[idx] = dn;
1462                 ctl->index++;
1463         } else {
1464                 dout("disable readdir cache\n");
1465                 ctl->index = -1;
1466         }
1467         return 0;
1468 }
1469
1470 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1471                              struct ceph_mds_session *session)
1472 {
1473         struct dentry *parent = req->r_dentry;
1474         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1475         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1476         struct qstr dname;
1477         struct dentry *dn;
1478         struct inode *in;
1479         int err = 0, skipped = 0, ret, i;
1480         struct inode *snapdir = NULL;
1481         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1482         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1483         u32 last_hash = 0;
1484         u32 fpos_offset;
1485         struct ceph_readdir_cache_control cache_ctl = {};
1486
1487         if (req->r_aborted)
1488                 return readdir_prepopulate_inodes_only(req, session);
1489
1490         if (rinfo->hash_order && req->r_path2) {
1491                 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1492                                           req->r_path2, strlen(req->r_path2));
1493                 last_hash = ceph_frag_value(last_hash);
1494         }
1495
1496         if (rinfo->dir_dir &&
1497             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1498                 dout("readdir_prepopulate got new frag %x -> %x\n",
1499                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1500                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1501                 if (!rinfo->hash_order)
1502                         req->r_readdir_offset = 2;
1503         }
1504
1505         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1506                 snapdir = ceph_get_snapdir(d_inode(parent));
1507                 parent = d_find_alias(snapdir);
1508                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1509                      rinfo->dir_nr, parent);
1510         } else {
1511                 dout("readdir_prepopulate %d items under dn %p\n",
1512                      rinfo->dir_nr, parent);
1513                 if (rinfo->dir_dir)
1514                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1515         }
1516
1517         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1518                 /* note dir version at start of readdir so we can tell
1519                  * if any dentries get dropped */
1520                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1521                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1522                 req->r_readdir_cache_idx = 0;
1523         }
1524
1525         cache_ctl.index = req->r_readdir_cache_idx;
1526         fpos_offset = req->r_readdir_offset;
1527
1528         /* FIXME: release caps/leases if error occurs */
1529         for (i = 0; i < rinfo->dir_nr; i++) {
1530                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1531                 struct ceph_vino vino;
1532
1533                 dname.name = rde->name;
1534                 dname.len = rde->name_len;
1535                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1536
1537                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1538                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1539
1540                 if (rinfo->hash_order) {
1541                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1542                                                  rde->name, rde->name_len);
1543                         hash = ceph_frag_value(hash);
1544                         if (hash != last_hash)
1545                                 fpos_offset = 2;
1546                         last_hash = hash;
1547                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1548                 } else {
1549                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1550                 }
1551
1552 retry_lookup:
1553                 dn = d_lookup(parent, &dname);
1554                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1555                      parent, dname.len, dname.name, dn);
1556
1557                 if (!dn) {
1558                         dn = d_alloc(parent, &dname);
1559                         dout("d_alloc %p '%.*s' = %p\n", parent,
1560                              dname.len, dname.name, dn);
1561                         if (dn == NULL) {
1562                                 dout("d_alloc badness\n");
1563                                 err = -ENOMEM;
1564                                 goto out;
1565                         }
1566                         ret = ceph_init_dentry(dn);
1567                         if (ret < 0) {
1568                                 dput(dn);
1569                                 err = ret;
1570                                 goto out;
1571                         }
1572                 } else if (d_really_is_positive(dn) &&
1573                            (ceph_ino(d_inode(dn)) != vino.ino ||
1574                             ceph_snap(d_inode(dn)) != vino.snap)) {
1575                         dout(" dn %p points to wrong inode %p\n",
1576                              dn, d_inode(dn));
1577                         d_delete(dn);
1578                         dput(dn);
1579                         goto retry_lookup;
1580                 }
1581
1582                 /* inode */
1583                 if (d_really_is_positive(dn)) {
1584                         in = d_inode(dn);
1585                 } else {
1586                         in = ceph_get_inode(parent->d_sb, vino);
1587                         if (IS_ERR(in)) {
1588                                 dout("new_inode badness\n");
1589                                 d_drop(dn);
1590                                 dput(dn);
1591                                 err = PTR_ERR(in);
1592                                 goto out;
1593                         }
1594                 }
1595
1596                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1597                                  req->r_request_started, -1,
1598                                  &req->r_caps_reservation);
1599                 if (ret < 0) {
1600                         pr_err("fill_inode badness on %p\n", in);
1601                         if (d_really_is_negative(dn))
1602                                 iput(in);
1603                         d_drop(dn);
1604                         err = ret;
1605                         goto next_item;
1606                 }
1607
1608                 if (d_really_is_negative(dn)) {
1609                         struct dentry *realdn;
1610
1611                         if (ceph_security_xattr_deadlock(in)) {
1612                                 dout(" skip splicing dn %p to inode %p"
1613                                      " (security xattr deadlock)\n", dn, in);
1614                                 iput(in);
1615                                 skipped++;
1616                                 goto next_item;
1617                         }
1618
1619                         realdn = splice_dentry(dn, in);
1620                         if (IS_ERR(realdn)) {
1621                                 err = PTR_ERR(realdn);
1622                                 d_drop(dn);
1623                                 dn = NULL;
1624                                 goto next_item;
1625                         }
1626                         dn = realdn;
1627                 }
1628
1629                 ceph_dentry(dn)->offset = rde->offset;
1630
1631                 update_dentry_lease(dn, rde->lease, req->r_session,
1632                                     req->r_request_started);
1633
1634                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1635                         ret = fill_readdir_cache(d_inode(parent), dn,
1636                                                  &cache_ctl, req);
1637                         if (ret < 0)
1638                                 err = ret;
1639                 }
1640 next_item:
1641                 if (dn)
1642                         dput(dn);
1643         }
1644 out:
1645         if (err == 0 && skipped == 0) {
1646                 req->r_did_prepopulate = true;
1647                 req->r_readdir_cache_idx = cache_ctl.index;
1648         }
1649         ceph_readdir_cache_release(&cache_ctl);
1650         if (snapdir) {
1651                 iput(snapdir);
1652                 dput(parent);
1653         }
1654         dout("readdir_prepopulate done\n");
1655         return err;
1656 }
1657
1658 int ceph_inode_set_size(struct inode *inode, loff_t size)
1659 {
1660         struct ceph_inode_info *ci = ceph_inode(inode);
1661         int ret = 0;
1662
1663         spin_lock(&ci->i_ceph_lock);
1664         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1665         i_size_write(inode, size);
1666         inode->i_blocks = calc_inode_blocks(size);
1667
1668         /* tell the MDS if we are approaching max_size */
1669         if ((size << 1) >= ci->i_max_size &&
1670             (ci->i_reported_size << 1) < ci->i_max_size)
1671                 ret = 1;
1672
1673         spin_unlock(&ci->i_ceph_lock);
1674         return ret;
1675 }
1676
1677 /*
1678  * Write back inode data in a worker thread.  (This can't be done
1679  * in the message handler context.)
1680  */
1681 void ceph_queue_writeback(struct inode *inode)
1682 {
1683         ihold(inode);
1684         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1685                        &ceph_inode(inode)->i_wb_work)) {
1686                 dout("ceph_queue_writeback %p\n", inode);
1687         } else {
1688                 dout("ceph_queue_writeback %p failed\n", inode);
1689                 iput(inode);
1690         }
1691 }
1692
1693 static void ceph_writeback_work(struct work_struct *work)
1694 {
1695         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1696                                                   i_wb_work);
1697         struct inode *inode = &ci->vfs_inode;
1698
1699         dout("writeback %p\n", inode);
1700         filemap_fdatawrite(&inode->i_data);
1701         iput(inode);
1702 }
1703
1704 /*
1705  * queue an async invalidation
1706  */
1707 void ceph_queue_invalidate(struct inode *inode)
1708 {
1709         ihold(inode);
1710         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1711                        &ceph_inode(inode)->i_pg_inv_work)) {
1712                 dout("ceph_queue_invalidate %p\n", inode);
1713         } else {
1714                 dout("ceph_queue_invalidate %p failed\n", inode);
1715                 iput(inode);
1716         }
1717 }
1718
1719 /*
1720  * Invalidate inode pages in a worker thread.  (This can't be done
1721  * in the message handler context.)
1722  */
1723 static void ceph_invalidate_work(struct work_struct *work)
1724 {
1725         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1726                                                   i_pg_inv_work);
1727         struct inode *inode = &ci->vfs_inode;
1728         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1729         u32 orig_gen;
1730         int check = 0;
1731
1732         mutex_lock(&ci->i_truncate_mutex);
1733
1734         if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1735                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1736                                     inode, ceph_ino(inode));
1737                 mapping_set_error(inode->i_mapping, -EIO);
1738                 truncate_pagecache(inode, 0);
1739                 mutex_unlock(&ci->i_truncate_mutex);
1740                 goto out;
1741         }
1742
1743         spin_lock(&ci->i_ceph_lock);
1744         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1745              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1746         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1747                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1748                         check = 1;
1749                 spin_unlock(&ci->i_ceph_lock);
1750                 mutex_unlock(&ci->i_truncate_mutex);
1751                 goto out;
1752         }
1753         orig_gen = ci->i_rdcache_gen;
1754         spin_unlock(&ci->i_ceph_lock);
1755
1756         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1757                 pr_err("invalidate_pages %p fails\n", inode);
1758         }
1759
1760         spin_lock(&ci->i_ceph_lock);
1761         if (orig_gen == ci->i_rdcache_gen &&
1762             orig_gen == ci->i_rdcache_revoking) {
1763                 dout("invalidate_pages %p gen %d successful\n", inode,
1764                      ci->i_rdcache_gen);
1765                 ci->i_rdcache_revoking--;
1766                 check = 1;
1767         } else {
1768                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1769                      inode, orig_gen, ci->i_rdcache_gen,
1770                      ci->i_rdcache_revoking);
1771                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1772                         check = 1;
1773         }
1774         spin_unlock(&ci->i_ceph_lock);
1775         mutex_unlock(&ci->i_truncate_mutex);
1776 out:
1777         if (check)
1778                 ceph_check_caps(ci, 0, NULL);
1779         iput(inode);
1780 }
1781
1782
1783 /*
1784  * called by trunc_wq;
1785  *
1786  * We also truncate in a separate thread as well.
1787  */
1788 static void ceph_vmtruncate_work(struct work_struct *work)
1789 {
1790         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1791                                                   i_vmtruncate_work);
1792         struct inode *inode = &ci->vfs_inode;
1793
1794         dout("vmtruncate_work %p\n", inode);
1795         __ceph_do_pending_vmtruncate(inode);
1796         iput(inode);
1797 }
1798
1799 /*
1800  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1801  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1802  */
1803 void ceph_queue_vmtruncate(struct inode *inode)
1804 {
1805         struct ceph_inode_info *ci = ceph_inode(inode);
1806
1807         ihold(inode);
1808
1809         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1810                        &ci->i_vmtruncate_work)) {
1811                 dout("ceph_queue_vmtruncate %p\n", inode);
1812         } else {
1813                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1814                      inode, ci->i_truncate_pending);
1815                 iput(inode);
1816         }
1817 }
1818
1819 /*
1820  * Make sure any pending truncation is applied before doing anything
1821  * that may depend on it.
1822  */
1823 void __ceph_do_pending_vmtruncate(struct inode *inode)
1824 {
1825         struct ceph_inode_info *ci = ceph_inode(inode);
1826         u64 to;
1827         int wrbuffer_refs, finish = 0;
1828
1829         mutex_lock(&ci->i_truncate_mutex);
1830 retry:
1831         spin_lock(&ci->i_ceph_lock);
1832         if (ci->i_truncate_pending == 0) {
1833                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1834                 spin_unlock(&ci->i_ceph_lock);
1835                 mutex_unlock(&ci->i_truncate_mutex);
1836                 return;
1837         }
1838
1839         /*
1840          * make sure any dirty snapped pages are flushed before we
1841          * possibly truncate them.. so write AND block!
1842          */
1843         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1844                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1845                      inode);
1846                 spin_unlock(&ci->i_ceph_lock);
1847                 filemap_write_and_wait_range(&inode->i_data, 0,
1848                                              inode->i_sb->s_maxbytes);
1849                 goto retry;
1850         }
1851
1852         /* there should be no reader or writer */
1853         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1854
1855         to = ci->i_truncate_size;
1856         wrbuffer_refs = ci->i_wrbuffer_ref;
1857         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1858              ci->i_truncate_pending, to);
1859         spin_unlock(&ci->i_ceph_lock);
1860
1861         truncate_pagecache(inode, to);
1862
1863         spin_lock(&ci->i_ceph_lock);
1864         if (to == ci->i_truncate_size) {
1865                 ci->i_truncate_pending = 0;
1866                 finish = 1;
1867         }
1868         spin_unlock(&ci->i_ceph_lock);
1869         if (!finish)
1870                 goto retry;
1871
1872         mutex_unlock(&ci->i_truncate_mutex);
1873
1874         if (wrbuffer_refs == 0)
1875                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1876
1877         wake_up_all(&ci->i_cap_wq);
1878 }
1879
1880 /*
1881  * symlinks
1882  */
1883 static const struct inode_operations ceph_symlink_iops = {
1884         .readlink = generic_readlink,
1885         .get_link = simple_get_link,
1886         .setattr = ceph_setattr,
1887         .getattr = ceph_getattr,
1888         .setxattr = generic_setxattr,
1889         .getxattr = generic_getxattr,
1890         .listxattr = ceph_listxattr,
1891         .removexattr = generic_removexattr,
1892 };
1893
1894 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1895 {
1896         struct ceph_inode_info *ci = ceph_inode(inode);
1897         const unsigned int ia_valid = attr->ia_valid;
1898         struct ceph_mds_request *req;
1899         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1900         struct ceph_cap_flush *prealloc_cf;
1901         int issued;
1902         int release = 0, dirtied = 0;
1903         int mask = 0;
1904         int err = 0;
1905         int inode_dirty_flags = 0;
1906         bool lock_snap_rwsem = false;
1907
1908         if (ceph_snap(inode) != CEPH_NOSNAP)
1909                 return -EROFS;
1910
1911         err = inode_change_ok(inode, attr);
1912         if (err != 0)
1913                 return err;
1914
1915         prealloc_cf = ceph_alloc_cap_flush();
1916         if (!prealloc_cf)
1917                 return -ENOMEM;
1918
1919         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1920                                        USE_AUTH_MDS);
1921         if (IS_ERR(req)) {
1922                 ceph_free_cap_flush(prealloc_cf);
1923                 return PTR_ERR(req);
1924         }
1925
1926         spin_lock(&ci->i_ceph_lock);
1927         issued = __ceph_caps_issued(ci, NULL);
1928
1929         if (!ci->i_head_snapc &&
1930             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1931                 lock_snap_rwsem = true;
1932                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1933                         spin_unlock(&ci->i_ceph_lock);
1934                         down_read(&mdsc->snap_rwsem);
1935                         spin_lock(&ci->i_ceph_lock);
1936                         issued = __ceph_caps_issued(ci, NULL);
1937                 }
1938         }
1939
1940         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1941
1942         if (ia_valid & ATTR_UID) {
1943                 dout("setattr %p uid %d -> %d\n", inode,
1944                      from_kuid(&init_user_ns, inode->i_uid),
1945                      from_kuid(&init_user_ns, attr->ia_uid));
1946                 if (issued & CEPH_CAP_AUTH_EXCL) {
1947                         inode->i_uid = attr->ia_uid;
1948                         dirtied |= CEPH_CAP_AUTH_EXCL;
1949                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1950                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1951                         req->r_args.setattr.uid = cpu_to_le32(
1952                                 from_kuid(&init_user_ns, attr->ia_uid));
1953                         mask |= CEPH_SETATTR_UID;
1954                         release |= CEPH_CAP_AUTH_SHARED;
1955                 }
1956         }
1957         if (ia_valid & ATTR_GID) {
1958                 dout("setattr %p gid %d -> %d\n", inode,
1959                      from_kgid(&init_user_ns, inode->i_gid),
1960                      from_kgid(&init_user_ns, attr->ia_gid));
1961                 if (issued & CEPH_CAP_AUTH_EXCL) {
1962                         inode->i_gid = attr->ia_gid;
1963                         dirtied |= CEPH_CAP_AUTH_EXCL;
1964                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1965                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1966                         req->r_args.setattr.gid = cpu_to_le32(
1967                                 from_kgid(&init_user_ns, attr->ia_gid));
1968                         mask |= CEPH_SETATTR_GID;
1969                         release |= CEPH_CAP_AUTH_SHARED;
1970                 }
1971         }
1972         if (ia_valid & ATTR_MODE) {
1973                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1974                      attr->ia_mode);
1975                 if (issued & CEPH_CAP_AUTH_EXCL) {
1976                         inode->i_mode = attr->ia_mode;
1977                         dirtied |= CEPH_CAP_AUTH_EXCL;
1978                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1979                            attr->ia_mode != inode->i_mode) {
1980                         inode->i_mode = attr->ia_mode;
1981                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1982                         mask |= CEPH_SETATTR_MODE;
1983                         release |= CEPH_CAP_AUTH_SHARED;
1984                 }
1985         }
1986
1987         if (ia_valid & ATTR_ATIME) {
1988                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1989                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1990                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1991                 if (issued & CEPH_CAP_FILE_EXCL) {
1992                         ci->i_time_warp_seq++;
1993                         inode->i_atime = attr->ia_atime;
1994                         dirtied |= CEPH_CAP_FILE_EXCL;
1995                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1996                            timespec_compare(&inode->i_atime,
1997                                             &attr->ia_atime) < 0) {
1998                         inode->i_atime = attr->ia_atime;
1999                         dirtied |= CEPH_CAP_FILE_WR;
2000                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2001                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
2002                         ceph_encode_timespec(&req->r_args.setattr.atime,
2003                                              &attr->ia_atime);
2004                         mask |= CEPH_SETATTR_ATIME;
2005                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
2006                                 CEPH_CAP_FILE_WR;
2007                 }
2008         }
2009         if (ia_valid & ATTR_MTIME) {
2010                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
2011                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2012                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2013                 if (issued & CEPH_CAP_FILE_EXCL) {
2014                         ci->i_time_warp_seq++;
2015                         inode->i_mtime = attr->ia_mtime;
2016                         dirtied |= CEPH_CAP_FILE_EXCL;
2017                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2018                            timespec_compare(&inode->i_mtime,
2019                                             &attr->ia_mtime) < 0) {
2020                         inode->i_mtime = attr->ia_mtime;
2021                         dirtied |= CEPH_CAP_FILE_WR;
2022                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2023                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2024                         ceph_encode_timespec(&req->r_args.setattr.mtime,
2025                                              &attr->ia_mtime);
2026                         mask |= CEPH_SETATTR_MTIME;
2027                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2028                                 CEPH_CAP_FILE_WR;
2029                 }
2030         }
2031         if (ia_valid & ATTR_SIZE) {
2032                 dout("setattr %p size %lld -> %lld\n", inode,
2033                      inode->i_size, attr->ia_size);
2034                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2035                     attr->ia_size > inode->i_size) {
2036                         i_size_write(inode, attr->ia_size);
2037                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2038                         inode->i_ctime = attr->ia_ctime;
2039                         ci->i_reported_size = attr->ia_size;
2040                         dirtied |= CEPH_CAP_FILE_EXCL;
2041                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2042                            attr->ia_size != inode->i_size) {
2043                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2044                         req->r_args.setattr.old_size =
2045                                 cpu_to_le64(inode->i_size);
2046                         mask |= CEPH_SETATTR_SIZE;
2047                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2048                                 CEPH_CAP_FILE_WR;
2049                 }
2050         }
2051
2052         /* these do nothing */
2053         if (ia_valid & ATTR_CTIME) {
2054                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2055                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2056                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2057                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2058                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2059                      only ? "ctime only" : "ignored");
2060                 inode->i_ctime = attr->ia_ctime;
2061                 if (only) {
2062                         /*
2063                          * if kernel wants to dirty ctime but nothing else,
2064                          * we need to choose a cap to dirty under, or do
2065                          * a almost-no-op setattr
2066                          */
2067                         if (issued & CEPH_CAP_AUTH_EXCL)
2068                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2069                         else if (issued & CEPH_CAP_FILE_EXCL)
2070                                 dirtied |= CEPH_CAP_FILE_EXCL;
2071                         else if (issued & CEPH_CAP_XATTR_EXCL)
2072                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2073                         else
2074                                 mask |= CEPH_SETATTR_CTIME;
2075                 }
2076         }
2077         if (ia_valid & ATTR_FILE)
2078                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2079
2080         if (dirtied) {
2081                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2082                                                            &prealloc_cf);
2083                 inode->i_ctime = current_fs_time(inode->i_sb);
2084         }
2085
2086         release &= issued;
2087         spin_unlock(&ci->i_ceph_lock);
2088         if (lock_snap_rwsem)
2089                 up_read(&mdsc->snap_rwsem);
2090
2091         if (inode_dirty_flags)
2092                 __mark_inode_dirty(inode, inode_dirty_flags);
2093
2094         if (ia_valid & ATTR_MODE) {
2095                 err = posix_acl_chmod(inode, attr->ia_mode);
2096                 if (err)
2097                         goto out_put;
2098         }
2099
2100         if (mask) {
2101                 req->r_inode = inode;
2102                 ihold(inode);
2103                 req->r_inode_drop = release;
2104                 req->r_args.setattr.mask = cpu_to_le32(mask);
2105                 req->r_num_caps = 1;
2106                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2107         }
2108         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2109              ceph_cap_string(dirtied), mask);
2110
2111         ceph_mdsc_put_request(req);
2112         if (mask & CEPH_SETATTR_SIZE)
2113                 __ceph_do_pending_vmtruncate(inode);
2114         ceph_free_cap_flush(prealloc_cf);
2115         return err;
2116 out_put:
2117         ceph_mdsc_put_request(req);
2118         ceph_free_cap_flush(prealloc_cf);
2119         return err;
2120 }
2121
2122 /*
2123  * setattr
2124  */
2125 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2126 {
2127         return __ceph_setattr(d_inode(dentry), attr);
2128 }
2129
2130 /*
2131  * Verify that we have a lease on the given mask.  If not,
2132  * do a getattr against an mds.
2133  */
2134 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2135                       int mask, bool force)
2136 {
2137         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2138         struct ceph_mds_client *mdsc = fsc->mdsc;
2139         struct ceph_mds_request *req;
2140         int err;
2141
2142         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2143                 dout("do_getattr inode %p SNAPDIR\n", inode);
2144                 return 0;
2145         }
2146
2147         dout("do_getattr inode %p mask %s mode 0%o\n",
2148              inode, ceph_cap_string(mask), inode->i_mode);
2149         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2150                 return 0;
2151
2152         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2153         if (IS_ERR(req))
2154                 return PTR_ERR(req);
2155         req->r_inode = inode;
2156         ihold(inode);
2157         req->r_num_caps = 1;
2158         req->r_args.getattr.mask = cpu_to_le32(mask);
2159         req->r_locked_page = locked_page;
2160         err = ceph_mdsc_do_request(mdsc, NULL, req);
2161         if (locked_page && err == 0) {
2162                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2163                 if (inline_version == 0) {
2164                         /* the reply is supposed to contain inline data */
2165                         err = -EINVAL;
2166                 } else if (inline_version == CEPH_INLINE_NONE) {
2167                         err = -ENODATA;
2168                 } else {
2169                         err = req->r_reply_info.targeti.inline_len;
2170                 }
2171         }
2172         ceph_mdsc_put_request(req);
2173         dout("do_getattr result=%d\n", err);
2174         return err;
2175 }
2176
2177
2178 /*
2179  * Check inode permissions.  We verify we have a valid value for
2180  * the AUTH cap, then call the generic handler.
2181  */
2182 int ceph_permission(struct inode *inode, int mask)
2183 {
2184         int err;
2185
2186         if (mask & MAY_NOT_BLOCK)
2187                 return -ECHILD;
2188
2189         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2190
2191         if (!err)
2192                 err = generic_permission(inode, mask);
2193         return err;
2194 }
2195
2196 /*
2197  * Get all attributes.  Hopefully somedata we'll have a statlite()
2198  * and can limit the fields we require to be accurate.
2199  */
2200 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2201                  struct kstat *stat)
2202 {
2203         struct inode *inode = d_inode(dentry);
2204         struct ceph_inode_info *ci = ceph_inode(inode);
2205         int err;
2206
2207         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2208         if (!err) {
2209                 generic_fillattr(inode, stat);
2210                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2211                 if (ceph_snap(inode) != CEPH_NOSNAP)
2212                         stat->dev = ceph_snap(inode);
2213                 else
2214                         stat->dev = 0;
2215                 if (S_ISDIR(inode->i_mode)) {
2216                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2217                                                 RBYTES))
2218                                 stat->size = ci->i_rbytes;
2219                         else
2220                                 stat->size = ci->i_files + ci->i_subdirs;
2221                         stat->blocks = 0;
2222                         stat->blksize = 65536;
2223                 }
2224         }
2225         return err;
2226 }