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         ci->i_pool_ns_len = 0;
 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         ci->i_cap_flush_tree = RB_ROOT;
 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_NUM; 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         call_rcu(&inode->i_rcu, ceph_i_callback);
 574 }
 575
 576 int ceph_drop_inode(struct inode *inode)
 577 {
 578         /*
 579          * Positve dentry and corresponding inode are always accompanied
 580          * in MDS reply. So no need to keep inode in the cache after
 581          * dropping all its aliases.
 582          */
 583         return 1;
 584 }
 585
 586 static inline blkcnt_t calc_inode_blocks(u64 size)
 587 {
 588         return (size + (1<<9) - 1) >> 9;
 589 }
 590
 591 /*
 592  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 593  * careful because either the client or MDS may have more up to date
 594  * info, depending on which capabilities are held, and whether
 595  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 596  * and size are monotonically increasing, except when utimes() or
 597  * truncate() increments the corresponding _seq values.)
 598  */
 599 int ceph_fill_file_size(struct inode *inode, int issued,
 600                         u32 truncate_seq, u64 truncate_size, u64 size)
 601 {
 602         struct ceph_inode_info *ci = ceph_inode(inode);
 603         int queue_trunc = 0;
 604
 605         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 606             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 607                 dout("size %lld -> %llu\n", inode->i_size, size);
 608                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 609                         pr_err("fill_file_size non-zero size for directory\n");
 610                         size = 0;
 611                 }
 612                 i_size_write(inode, size);
 613                 inode->i_blocks = calc_inode_blocks(size);
 614                 ci->i_reported_size = size;
 615                 if (truncate_seq != ci->i_truncate_seq) {
 616                         dout("truncate_seq %u -> %u\n",
 617                              ci->i_truncate_seq, truncate_seq);
 618                         ci->i_truncate_seq = truncate_seq;
 619
 620                         /* the MDS should have revoked these caps */
 621                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 622                                                CEPH_CAP_FILE_RD |
 623                                                CEPH_CAP_FILE_WR |
 624                                                CEPH_CAP_FILE_LAZYIO));
 625                         /*
 626                          * If we hold relevant caps, or in the case where we're
 627                          * not the only client referencing this file and we
 628                          * don't hold those caps, then we need to check whether
 629                          * the file is either opened or mmaped
 630                          */
 631                         if ((issued & (CEPH_CAP_FILE_CACHE|
 632                                        CEPH_CAP_FILE_BUFFER)) ||
 633                             mapping_mapped(inode->i_mapping) ||
 634                             __ceph_caps_file_wanted(ci)) {
 635                                 ci->i_truncate_pending++;
 636                                 queue_trunc = 1;
 637                         }
 638                 }
 639         }
 640         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 641             ci->i_truncate_size != truncate_size) {
 642                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 643                      truncate_size);
 644                 ci->i_truncate_size = truncate_size;
 645         }
 646
 647         if (queue_trunc)
 648                 ceph_fscache_invalidate(inode);
 649
 650         return queue_trunc;
 651 }
 652
 653 void ceph_fill_file_time(struct inode *inode, int issued,
 654                          u64 time_warp_seq, struct timespec *ctime,
 655                          struct timespec *mtime, struct timespec *atime)
 656 {
 657         struct ceph_inode_info *ci = ceph_inode(inode);
 658         int warn = 0;
 659
 660         if (issued & (CEPH_CAP_FILE_EXCL|
 661                       CEPH_CAP_FILE_WR|
 662                       CEPH_CAP_FILE_BUFFER|
 663                       CEPH_CAP_AUTH_EXCL|
 664                       CEPH_CAP_XATTR_EXCL)) {
 665                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 666                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 667                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 668                              ctime->tv_sec, ctime->tv_nsec);
 669                         inode->i_ctime = *ctime;
 670                 }
 671                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 672                         /* the MDS did a utimes() */
 673                         dout("mtime %ld.%09ld -> %ld.%09ld "
 674                              "tw %d -> %d\n",
 675                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 676                              mtime->tv_sec, mtime->tv_nsec,
 677                              ci->i_time_warp_seq, (int)time_warp_seq);
 678
 679                         inode->i_mtime = *mtime;
 680                         inode->i_atime = *atime;
 681                         ci->i_time_warp_seq = time_warp_seq;
 682                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 683                         /* nobody did utimes(); take the max */
 684                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 685                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 686                                      inode->i_mtime.tv_sec,
 687                                      inode->i_mtime.tv_nsec,
 688                                      mtime->tv_sec, mtime->tv_nsec);
 689                                 inode->i_mtime = *mtime;
 690                         }
 691                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 692                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 693                                      inode->i_atime.tv_sec,
 694                                      inode->i_atime.tv_nsec,
 695                                      atime->tv_sec, atime->tv_nsec);
 696                                 inode->i_atime = *atime;
 697                         }
 698                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 699                         /* we did a utimes(); ignore mds values */
 700                 } else {
 701                         warn = 1;
 702                 }
 703         } else {
 704                 /* we have no write|excl caps; whatever the MDS says is true */
 705                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 706                         inode->i_ctime = *ctime;
 707                         inode->i_mtime = *mtime;
 708                         inode->i_atime = *atime;
 709                         ci->i_time_warp_seq = time_warp_seq;
 710                 } else {
 711                         warn = 1;
 712                 }
 713         }
 714         if (warn) /* time_warp_seq shouldn't go backwards */
 715                 dout("%p mds time_warp_seq %llu < %u\n",
 716                      inode, time_warp_seq, ci->i_time_warp_seq);
 717 }
 718
 719 /*
 720  * Populate an inode based on info from mds.  May be called on new or
 721  * existing inodes.
 722  */
 723 static int fill_inode(struct inode *inode, struct page *locked_page,
 724                       struct ceph_mds_reply_info_in *iinfo,
 725                       struct ceph_mds_reply_dirfrag *dirinfo,
 726                       struct ceph_mds_session *session,
 727                       unsigned long ttl_from, int cap_fmode,
 728                       struct ceph_cap_reservation *caps_reservation)
 729 {
 730         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 731         struct ceph_mds_reply_inode *info = iinfo->in;
 732         struct ceph_inode_info *ci = ceph_inode(inode);
 733         int issued = 0, implemented, new_issued;
 734         struct timespec mtime, atime, ctime;
 735         struct ceph_buffer *xattr_blob = NULL;
 736         struct ceph_cap *new_cap = NULL;
 737         int err = 0;
 738         bool wake = false;
 739         bool queue_trunc = false;
 740         bool new_version = false;
 741         bool fill_inline = false;
 742
 743         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 744              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 745              ci->i_version);
 746
 747         /* prealloc new cap struct */
 748         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 749                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 750
 751         /*
 752          * prealloc xattr data, if it looks like we'll need it.  only
 753          * if len > 4 (meaning there are actually xattrs; the first 4
 754          * bytes are the xattr count).
 755          */
 756         if (iinfo->xattr_len > 4) {
 757                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 758                 if (!xattr_blob)
 759                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 760                                iinfo->xattr_len);
 761         }
 762
 763         spin_lock(&ci->i_ceph_lock);
 764
 765         /*
 766          * provided version will be odd if inode value is projected,
 767          * even if stable.  skip the update if we have newer stable
 768          * info (ours>=theirs, e.g. due to racing mds replies), unless
 769          * we are getting projected (unstable) info (in which case the
 770          * version is odd, and we want ours>theirs).
 771          *   us   them
 772          *   2    2     skip
 773          *   3    2     skip
 774          *   3    3     update
 775          */
 776         if (ci->i_version == 0 ||
 777             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 778              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 779                 new_version = true;
 780
 781         issued = __ceph_caps_issued(ci, &implemented);
 782         issued |= implemented | __ceph_caps_dirty(ci);
 783         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 784
 785         /* update inode */
 786         ci->i_version = le64_to_cpu(info->version);
 787         inode->i_version++;
 788         inode->i_rdev = le32_to_cpu(info->rdev);
 789         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 790
 791         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 792             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 793                 inode->i_mode = le32_to_cpu(info->mode);
 794                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 795                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 796                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 797                      from_kuid(&init_user_ns, inode->i_uid),
 798                      from_kgid(&init_user_ns, inode->i_gid));
 799         }
 800
 801         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 802             (issued & CEPH_CAP_LINK_EXCL) == 0)
 803                 set_nlink(inode, le32_to_cpu(info->nlink));
 804
 805         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 806                 /* be careful with mtime, atime, size */
 807                 ceph_decode_timespec(&atime, &info->atime);
 808                 ceph_decode_timespec(&mtime, &info->mtime);
 809                 ceph_decode_timespec(&ctime, &info->ctime);
 810                 ceph_fill_file_time(inode, issued,
 811                                 le32_to_cpu(info->time_warp_seq),
 812                                 &ctime, &mtime, &atime);
 813         }
 814
 815         if (new_version ||
 816             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 817                 s64 old_pool = ci->i_layout.pool_id;
 818                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 819                 ci->i_pool_ns_len = iinfo->pool_ns_len;
 820                 if (ci->i_layout.pool_id != old_pool)
 821                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 822
 823                 queue_trunc = ceph_fill_file_size(inode, issued,
 824                                         le32_to_cpu(info->truncate_seq),
 825                                         le64_to_cpu(info->truncate_size),
 826                                         le64_to_cpu(info->size));
 827                 /* only update max_size on auth cap */
 828                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 829                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 830                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 831                                         le64_to_cpu(info->max_size));
 832                         ci->i_max_size = le64_to_cpu(info->max_size);
 833                 }
 834         }
 835
 836         /* xattrs */
 837         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 838         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 839             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 840                 if (ci->i_xattrs.blob)
 841                         ceph_buffer_put(ci->i_xattrs.blob);
 842                 ci->i_xattrs.blob = xattr_blob;
 843                 if (xattr_blob)
 844                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 845                                iinfo->xattr_data, iinfo->xattr_len);
 846                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 847                 ceph_forget_all_cached_acls(inode);
 848                 xattr_blob = NULL;
 849         }
 850
 851         inode->i_mapping->a_ops = &ceph_aops;
 852
 853         switch (inode->i_mode & S_IFMT) {
 854         case S_IFIFO:
 855         case S_IFBLK:
 856         case S_IFCHR:
 857         case S_IFSOCK:
 858                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 859                 inode->i_op = &ceph_file_iops;
 860                 break;
 861         case S_IFREG:
 862                 inode->i_op = &ceph_file_iops;
 863                 inode->i_fop = &ceph_file_fops;
 864                 break;
 865         case S_IFLNK:
 866                 inode->i_op = &ceph_symlink_iops;
 867                 if (!ci->i_symlink) {
 868                         u32 symlen = iinfo->symlink_len;
 869                         char *sym;
 870
 871                         spin_unlock(&ci->i_ceph_lock);
 872
 873                         if (symlen != i_size_read(inode)) {
 874                                 pr_err("fill_inode %llx.%llx BAD symlink "
 875                                         "size %lld\n", ceph_vinop(inode),
 876                                         i_size_read(inode));
 877                                 i_size_write(inode, symlen);
 878                                 inode->i_blocks = calc_inode_blocks(symlen);
 879                         }
 880
 881                         err = -ENOMEM;
 882                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 883                         if (!sym)
 884                                 goto out;
 885
 886                         spin_lock(&ci->i_ceph_lock);
 887                         if (!ci->i_symlink)
 888                                 ci->i_symlink = sym;
 889                         else
 890                                 kfree(sym); /* lost a race */
 891                 }
 892                 inode->i_link = ci->i_symlink;
 893                 break;
 894         case S_IFDIR:
 895                 inode->i_op = &ceph_dir_iops;
 896                 inode->i_fop = &ceph_dir_fops;
 897
 898                 ci->i_dir_layout = iinfo->dir_layout;
 899
 900                 ci->i_files = le64_to_cpu(info->files);
 901                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 902                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 903                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 904                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 905                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 906                 break;
 907         default:
 908                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 909                        ceph_vinop(inode), inode->i_mode);
 910         }
 911
 912         /* were we issued a capability? */
 913         if (info->cap.caps) {
 914                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 915                         unsigned caps = le32_to_cpu(info->cap.caps);
 916                         ceph_add_cap(inode, session,
 917                                      le64_to_cpu(info->cap.cap_id),
 918                                      cap_fmode, caps,
 919                                      le32_to_cpu(info->cap.wanted),
 920                                      le32_to_cpu(info->cap.seq),
 921                                      le32_to_cpu(info->cap.mseq),
 922                                      le64_to_cpu(info->cap.realm),
 923                                      info->cap.flags, &new_cap);
 924
 925                         /* set dir completion flag? */
 926                         if (S_ISDIR(inode->i_mode) &&
 927                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 928                             (caps & CEPH_CAP_FILE_SHARED) &&
 929                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 930                             !__ceph_dir_is_complete(ci)) {
 931                                 dout(" marking %p complete (empty)\n", inode);
 932                                 i_size_write(inode, 0);
 933                                 __ceph_dir_set_complete(ci,
 934                                         atomic64_read(&ci->i_release_count),
 935                                         atomic64_read(&ci->i_ordered_count));
 936                         }
 937
 938                         wake = true;
 939                 } else {
 940                         dout(" %p got snap_caps %s\n", inode,
 941                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 942                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 943                         if (cap_fmode >= 0)
 944                                 __ceph_get_fmode(ci, cap_fmode);
 945                 }
 946         } else if (cap_fmode >= 0) {
 947                 pr_warn("mds issued no caps on %llx.%llx\n",
 948                            ceph_vinop(inode));
 949                 __ceph_get_fmode(ci, cap_fmode);
 950         }
 951
 952         if (iinfo->inline_version > 0 &&
 953             iinfo->inline_version >= ci->i_inline_version) {
 954                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 955                 ci->i_inline_version = iinfo->inline_version;
 956                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 957                     (locked_page ||
 958                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 959                         fill_inline = true;
 960         }
 961
 962         spin_unlock(&ci->i_ceph_lock);
 963
 964         if (fill_inline)
 965                 ceph_fill_inline_data(inode, locked_page,
 966                                       iinfo->inline_data, iinfo->inline_len);
 967
 968         if (wake)
 969                 wake_up_all(&ci->i_cap_wq);
 970
 971         /* queue truncate if we saw i_size decrease */
 972         if (queue_trunc)
 973                 ceph_queue_vmtruncate(inode);
 974
 975         /* populate frag tree */
 976         if (S_ISDIR(inode->i_mode))
 977                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 978
 979         /* update delegation info? */
 980         if (dirinfo)
 981                 ceph_fill_dirfrag(inode, dirinfo);
 982
 983         err = 0;
 984 out:
 985         if (new_cap)
 986                 ceph_put_cap(mdsc, new_cap);
 987         if (xattr_blob)
 988                 ceph_buffer_put(xattr_blob);
 989         return err;
 990 }
 991
 992 /*
 993  * caller should hold session s_mutex.
 994  */
 995 static void update_dentry_lease(struct dentry *dentry,
 996                                 struct ceph_mds_reply_lease *lease,
 997                                 struct ceph_mds_session *session,
 998                                 unsigned long from_time)
 999 {
1000         struct ceph_dentry_info *di = ceph_dentry(dentry);
1001         long unsigned duration = le32_to_cpu(lease->duration_ms);
1002         long unsigned ttl = from_time + (duration * HZ) / 1000;
1003         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1004         struct inode *dir;
1005
1006         /* only track leases on regular dentries */
1007         if (dentry->d_op != &ceph_dentry_ops)
1008                 return;
1009
1010         spin_lock(&dentry->d_lock);
1011         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1012              dentry, duration, ttl);
1013
1014         /* make lease_rdcache_gen match directory */
1015         dir = d_inode(dentry->d_parent);
1016         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1017
1018         if (duration == 0)
1019                 goto out_unlock;
1020
1021         if (di->lease_gen == session->s_cap_gen &&
1022             time_before(ttl, dentry->d_time))
1023                 goto out_unlock;  /* we already have a newer lease. */
1024
1025         if (di->lease_session && di->lease_session != session)
1026                 goto out_unlock;
1027
1028         ceph_dentry_lru_touch(dentry);
1029
1030         if (!di->lease_session)
1031                 di->lease_session = ceph_get_mds_session(session);
1032         di->lease_gen = session->s_cap_gen;
1033         di->lease_seq = le32_to_cpu(lease->seq);
1034         di->lease_renew_after = half_ttl;
1035         di->lease_renew_from = 0;
1036         dentry->d_time = ttl;
1037 out_unlock:
1038         spin_unlock(&dentry->d_lock);
1039         return;
1040 }
1041
1042 /*
1043  * splice a dentry to an inode.
1044  * caller must hold directory i_mutex for this to be safe.
1045  */
1046 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1047 {
1048         struct dentry *realdn;
1049
1050         BUG_ON(d_inode(dn));
1051
1052         /* dn must be unhashed */
1053         if (!d_unhashed(dn))
1054                 d_drop(dn);
1055         realdn = d_splice_alias(in, dn);
1056         if (IS_ERR(realdn)) {
1057                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1058                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1059                 dn = realdn; /* note realdn contains the error */
1060                 goto out;
1061         } else if (realdn) {
1062                 dout("dn %p (%d) spliced with %p (%d) "
1063                      "inode %p ino %llx.%llx\n",
1064                      dn, d_count(dn),
1065                      realdn, d_count(realdn),
1066                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1067                 dput(dn);
1068                 dn = realdn;
1069         } else {
1070                 BUG_ON(!ceph_dentry(dn));
1071                 dout("dn %p attached to %p ino %llx.%llx\n",
1072                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1073         }
1074 out:
1075         return dn;
1076 }
1077
1078 /*
1079  * Incorporate results into the local cache.  This is either just
1080  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1081  * after a lookup).
1082  *
1083  * A reply may contain
1084  *         a directory inode along with a dentry.
1085  *  and/or a target inode
1086  *
1087  * Called with snap_rwsem (read).
1088  */
1089 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1090                     struct ceph_mds_session *session)
1091 {
1092         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1093         struct inode *in = NULL;
1094         struct ceph_vino vino;
1095         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1096         int err = 0;
1097
1098         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1099              rinfo->head->is_dentry, rinfo->head->is_target);
1100
1101 #if 0
1102         /*
1103          * Debugging hook:
1104          *
1105          * If we resend completed ops to a recovering mds, we get no
1106          * trace.  Since that is very rare, pretend this is the case
1107          * to ensure the 'no trace' handlers in the callers behave.
1108          *
1109          * Fill in inodes unconditionally to avoid breaking cap
1110          * invariants.
1111          */
1112         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1113                 pr_info("fill_trace faking empty trace on %lld %s\n",
1114                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1115                 if (rinfo->head->is_dentry) {
1116                         rinfo->head->is_dentry = 0;
1117                         err = fill_inode(req->r_locked_dir,
1118                                          &rinfo->diri, rinfo->dirfrag,
1119                                          session, req->r_request_started, -1);
1120                 }
1121                 if (rinfo->head->is_target) {
1122                         rinfo->head->is_target = 0;
1123                         ininfo = rinfo->targeti.in;
1124                         vino.ino = le64_to_cpu(ininfo->ino);
1125                         vino.snap = le64_to_cpu(ininfo->snapid);
1126                         in = ceph_get_inode(sb, vino);
1127                         err = fill_inode(in, &rinfo->targeti, NULL,
1128                                          session, req->r_request_started,
1129                                          req->r_fmode);
1130                         iput(in);
1131                 }
1132         }
1133 #endif
1134
1135         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1136                 dout("fill_trace reply is empty!\n");
1137                 if (rinfo->head->result == 0 && req->r_locked_dir)
1138                         ceph_invalidate_dir_request(req);
1139                 return 0;
1140         }
1141
1142         if (rinfo->head->is_dentry) {
1143                 struct inode *dir = req->r_locked_dir;
1144
1145                 if (dir) {
1146                         err = fill_inode(dir, NULL,
1147                                          &rinfo->diri, rinfo->dirfrag,
1148                                          session, req->r_request_started, -1,
1149                                          &req->r_caps_reservation);
1150                         if (err < 0)
1151                                 goto done;
1152                 } else {
1153                         WARN_ON_ONCE(1);
1154                 }
1155
1156                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1157                         struct qstr dname;
1158                         struct dentry *dn, *parent;
1159
1160                         BUG_ON(!rinfo->head->is_target);
1161                         BUG_ON(req->r_dentry);
1162
1163                         parent = d_find_any_alias(dir);
1164                         BUG_ON(!parent);
1165
1166                         dname.name = rinfo->dname;
1167                         dname.len = rinfo->dname_len;
1168                         dname.hash = full_name_hash(dname.name, dname.len);
1169                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1170                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1171 retry_lookup:
1172                         dn = d_lookup(parent, &dname);
1173                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1174                              parent, dname.len, dname.name, dn);
1175
1176                         if (!dn) {
1177                                 dn = d_alloc(parent, &dname);
1178                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1179                                      dname.len, dname.name, dn);
1180                                 if (dn == NULL) {
1181                                         dput(parent);
1182                                         err = -ENOMEM;
1183                                         goto done;
1184                                 }
1185                                 err = ceph_init_dentry(dn);
1186                                 if (err < 0) {
1187                                         dput(dn);
1188                                         dput(parent);
1189                                         goto done;
1190                                 }
1191                         } else if (d_really_is_positive(dn) &&
1192                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1193                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1194                                 dout(" dn %p points to wrong inode %p\n",
1195                                      dn, d_inode(dn));
1196                                 d_delete(dn);
1197                                 dput(dn);
1198                                 goto retry_lookup;
1199                         }
1200
1201                         req->r_dentry = dn;
1202                         dput(parent);
1203                 }
1204         }
1205
1206         if (rinfo->head->is_target) {
1207                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1208                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1209
1210                 in = ceph_get_inode(sb, vino);
1211                 if (IS_ERR(in)) {
1212                         err = PTR_ERR(in);
1213                         goto done;
1214                 }
1215                 req->r_target_inode = in;
1216
1217                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1218                                 session, req->r_request_started,
1219                                 (!req->r_aborted && rinfo->head->result == 0) ?
1220                                 req->r_fmode : -1,
1221                                 &req->r_caps_reservation);
1222                 if (err < 0) {
1223                         pr_err("fill_inode badness %p %llx.%llx\n",
1224                                 in, ceph_vinop(in));
1225                         goto done;
1226                 }
1227         }
1228
1229         /*
1230          * ignore null lease/binding on snapdir ENOENT, or else we
1231          * will have trouble splicing in the virtual snapdir later
1232          */
1233         if (rinfo->head->is_dentry && !req->r_aborted &&
1234             req->r_locked_dir &&
1235             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1236                                                fsc->mount_options->snapdir_name,
1237                                                req->r_dentry->d_name.len))) {
1238                 /*
1239                  * lookup link rename   : null -> possibly existing inode
1240                  * mknod symlink mkdir  : null -> new inode
1241                  * unlink               : linked -> null
1242                  */
1243                 struct inode *dir = req->r_locked_dir;
1244                 struct dentry *dn = req->r_dentry;
1245                 bool have_dir_cap, have_lease;
1246
1247                 BUG_ON(!dn);
1248                 BUG_ON(!dir);
1249                 BUG_ON(d_inode(dn->d_parent) != dir);
1250                 BUG_ON(ceph_ino(dir) !=
1251                        le64_to_cpu(rinfo->diri.in->ino));
1252                 BUG_ON(ceph_snap(dir) !=
1253                        le64_to_cpu(rinfo->diri.in->snapid));
1254
1255                 /* do we have a lease on the whole dir? */
1256                 have_dir_cap =
1257                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1258                          CEPH_CAP_FILE_SHARED);
1259
1260                 /* do we have a dn lease? */
1261                 have_lease = have_dir_cap ||
1262                         le32_to_cpu(rinfo->dlease->duration_ms);
1263                 if (!have_lease)
1264                         dout("fill_trace  no dentry lease or dir cap\n");
1265
1266                 /* rename? */
1267                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1268                         struct inode *olddir = req->r_old_dentry_dir;
1269                         BUG_ON(!olddir);
1270
1271                         dout(" src %p '%pd' dst %p '%pd'\n",
1272                              req->r_old_dentry,
1273                              req->r_old_dentry,
1274                              dn, dn);
1275                         dout("fill_trace doing d_move %p -> %p\n",
1276                              req->r_old_dentry, dn);
1277
1278                         /* d_move screws up sibling dentries' offsets */
1279                         ceph_dir_clear_ordered(dir);
1280                         ceph_dir_clear_ordered(olddir);
1281
1282                         d_move(req->r_old_dentry, dn);
1283                         dout(" src %p '%pd' dst %p '%pd'\n",
1284                              req->r_old_dentry,
1285                              req->r_old_dentry,
1286                              dn, dn);
1287
1288                         /* ensure target dentry is invalidated, despite
1289                            rehashing bug in vfs_rename_dir */
1290                         ceph_invalidate_dentry_lease(dn);
1291
1292                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1293                              ceph_dentry(req->r_old_dentry)->offset);
1294
1295                         dn = req->r_old_dentry;  /* use old_dentry */
1296                 }
1297
1298                 /* null dentry? */
1299                 if (!rinfo->head->is_target) {
1300                         dout("fill_trace null dentry\n");
1301                         if (d_really_is_positive(dn)) {
1302                                 ceph_dir_clear_ordered(dir);
1303                                 dout("d_delete %p\n", dn);
1304                                 d_delete(dn);
1305                         } else {
1306                                 if (have_lease && d_unhashed(dn))
1307                                         d_add(dn, NULL);
1308                                 update_dentry_lease(dn, rinfo->dlease,
1309                                                     session,
1310                                                     req->r_request_started);
1311                         }
1312                         goto done;
1313                 }
1314
1315                 /* attach proper inode */
1316                 if (d_really_is_negative(dn)) {
1317                         ceph_dir_clear_ordered(dir);
1318                         ihold(in);
1319                         dn = splice_dentry(dn, in);
1320                         if (IS_ERR(dn)) {
1321                                 err = PTR_ERR(dn);
1322                                 goto done;
1323                         }
1324                         req->r_dentry = dn;  /* may have spliced */
1325                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1326                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1327                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1328                              ceph_vinop(in));
1329                         d_invalidate(dn);
1330                         have_lease = false;
1331                 }
1332
1333                 if (have_lease)
1334                         update_dentry_lease(dn, rinfo->dlease, session,
1335                                             req->r_request_started);
1336                 dout(" final dn %p\n", dn);
1337         } else if (!req->r_aborted &&
1338                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1339                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1340                 struct dentry *dn = req->r_dentry;
1341                 struct inode *dir = req->r_locked_dir;
1342
1343                 /* fill out a snapdir LOOKUPSNAP dentry */
1344                 BUG_ON(!dn);
1345                 BUG_ON(!dir);
1346                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1347                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1348                 ceph_dir_clear_ordered(dir);
1349                 ihold(in);
1350                 dn = splice_dentry(dn, in);
1351                 if (IS_ERR(dn)) {
1352                         err = PTR_ERR(dn);
1353                         goto done;
1354                 }
1355                 req->r_dentry = dn;  /* may have spliced */
1356         }
1357 done:
1358         dout("fill_trace done err=%d\n", err);
1359         return err;
1360 }
1361
1362 /*
1363  * Prepopulate our cache with readdir results, leases, etc.
1364  */
1365 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1366                                            struct ceph_mds_session *session)
1367 {
1368         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1369         int i, err = 0;
1370
1371         for (i = 0; i < rinfo->dir_nr; i++) {
1372                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1373                 struct ceph_vino vino;
1374                 struct inode *in;
1375                 int rc;
1376
1377                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1378                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1379
1380                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1381                 if (IS_ERR(in)) {
1382                         err = PTR_ERR(in);
1383                         dout("new_inode badness got %d\n", err);
1384                         continue;
1385                 }
1386                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1387                                 req->r_request_started, -1,
1388                                 &req->r_caps_reservation);
1389                 if (rc < 0) {
1390                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1391                         err = rc;
1392                 }
1393                 iput(in);
1394         }
1395
1396         return err;
1397 }
1398
1399 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1400 {
1401         if (ctl->page) {
1402                 kunmap(ctl->page);
1403                 put_page(ctl->page);
1404                 ctl->page = NULL;
1405         }
1406 }
1407
1408 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1409                               struct ceph_readdir_cache_control *ctl,
1410                               struct ceph_mds_request *req)
1411 {
1412         struct ceph_inode_info *ci = ceph_inode(dir);
1413         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1414         unsigned idx = ctl->index % nsize;
1415         pgoff_t pgoff = ctl->index / nsize;
1416
1417         if (!ctl->page || pgoff != page_index(ctl->page)) {
1418                 ceph_readdir_cache_release(ctl);
1419                 if (idx == 0)
1420                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1421                 else
1422                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1423                 if (!ctl->page) {
1424                         ctl->index = -1;
1425                         return idx == 0 ? -ENOMEM : 0;
1426                 }
1427                 /* reading/filling the cache are serialized by
1428                  * i_mutex, no need to use page lock */
1429                 unlock_page(ctl->page);
1430                 ctl->dentries = kmap(ctl->page);
1431                 if (idx == 0)
1432                         memset(ctl->dentries, 0, PAGE_SIZE);
1433         }
1434
1435         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1436             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1437                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1438                 ctl->dentries[idx] = dn;
1439                 ctl->index++;
1440         } else {
1441                 dout("disable readdir cache\n");
1442                 ctl->index = -1;
1443         }
1444         return 0;
1445 }
1446
1447 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1448                              struct ceph_mds_session *session)
1449 {
1450         struct dentry *parent = req->r_dentry;
1451         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1452         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1453         struct qstr dname;
1454         struct dentry *dn;
1455         struct inode *in;
1456         int err = 0, skipped = 0, ret, i;
1457         struct inode *snapdir = NULL;
1458         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1459         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1460         u32 last_hash = 0;
1461         u32 fpos_offset;
1462         struct ceph_readdir_cache_control cache_ctl = {};
1463
1464         if (req->r_aborted)
1465                 return readdir_prepopulate_inodes_only(req, session);
1466
1467         if (rinfo->hash_order && req->r_path2) {
1468                 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1469                                           req->r_path2, strlen(req->r_path2));
1470                 last_hash = ceph_frag_value(last_hash);
1471         }
1472
1473         if (rinfo->dir_dir &&
1474             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1475                 dout("readdir_prepopulate got new frag %x -> %x\n",
1476                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1477                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1478                 if (!rinfo->hash_order)
1479                         req->r_readdir_offset = 2;
1480         }
1481
1482         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1483                 snapdir = ceph_get_snapdir(d_inode(parent));
1484                 parent = d_find_alias(snapdir);
1485                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1486                      rinfo->dir_nr, parent);
1487         } else {
1488                 dout("readdir_prepopulate %d items under dn %p\n",
1489                      rinfo->dir_nr, parent);
1490                 if (rinfo->dir_dir)
1491                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1492         }
1493
1494         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1495                 /* note dir version at start of readdir so we can tell
1496                  * if any dentries get dropped */
1497                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1498                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1499                 req->r_readdir_cache_idx = 0;
1500         }
1501
1502         cache_ctl.index = req->r_readdir_cache_idx;
1503         fpos_offset = req->r_readdir_offset;
1504
1505         /* FIXME: release caps/leases if error occurs */
1506         for (i = 0; i < rinfo->dir_nr; i++) {
1507                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1508                 struct ceph_vino vino;
1509
1510                 dname.name = rde->name;
1511                 dname.len = rde->name_len;
1512                 dname.hash = full_name_hash(dname.name, dname.len);
1513
1514                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1515                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1516
1517                 if (rinfo->hash_order) {
1518                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1519                                                  rde->name, rde->name_len);
1520                         hash = ceph_frag_value(hash);
1521                         if (hash != last_hash)
1522                                 fpos_offset = 2;
1523                         last_hash = hash;
1524                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1525                 } else {
1526                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1527                 }
1528
1529 retry_lookup:
1530                 dn = d_lookup(parent, &dname);
1531                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1532                      parent, dname.len, dname.name, dn);
1533
1534                 if (!dn) {
1535                         dn = d_alloc(parent, &dname);
1536                         dout("d_alloc %p '%.*s' = %p\n", parent,
1537                              dname.len, dname.name, dn);
1538                         if (dn == NULL) {
1539                                 dout("d_alloc badness\n");
1540                                 err = -ENOMEM;
1541                                 goto out;
1542                         }
1543                         ret = ceph_init_dentry(dn);
1544                         if (ret < 0) {
1545                                 dput(dn);
1546                                 err = ret;
1547                                 goto out;
1548                         }
1549                 } else if (d_really_is_positive(dn) &&
1550                            (ceph_ino(d_inode(dn)) != vino.ino ||
1551                             ceph_snap(d_inode(dn)) != vino.snap)) {
1552                         dout(" dn %p points to wrong inode %p\n",
1553                              dn, d_inode(dn));
1554                         d_delete(dn);
1555                         dput(dn);
1556                         goto retry_lookup;
1557                 }
1558
1559                 /* inode */
1560                 if (d_really_is_positive(dn)) {
1561                         in = d_inode(dn);
1562                 } else {
1563                         in = ceph_get_inode(parent->d_sb, vino);
1564                         if (IS_ERR(in)) {
1565                                 dout("new_inode badness\n");
1566                                 d_drop(dn);
1567                                 dput(dn);
1568                                 err = PTR_ERR(in);
1569                                 goto out;
1570                         }
1571                 }
1572
1573                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1574                                  req->r_request_started, -1,
1575                                  &req->r_caps_reservation);
1576                 if (ret < 0) {
1577                         pr_err("fill_inode badness on %p\n", in);
1578                         if (d_really_is_negative(dn))
1579                                 iput(in);
1580                         d_drop(dn);
1581                         err = ret;
1582                         goto next_item;
1583                 }
1584
1585                 if (d_really_is_negative(dn)) {
1586                         struct dentry *realdn;
1587
1588                         if (ceph_security_xattr_deadlock(in)) {
1589                                 dout(" skip splicing dn %p to inode %p"
1590                                      " (security xattr deadlock)\n", dn, in);
1591                                 iput(in);
1592                                 skipped++;
1593                                 goto next_item;
1594                         }
1595
1596                         realdn = splice_dentry(dn, in);
1597                         if (IS_ERR(realdn)) {
1598                                 err = PTR_ERR(realdn);
1599                                 d_drop(dn);
1600                                 dn = NULL;
1601                                 goto next_item;
1602                         }
1603                         dn = realdn;
1604                 }
1605
1606                 ceph_dentry(dn)->offset = rde->offset;
1607
1608                 update_dentry_lease(dn, rde->lease, req->r_session,
1609                                     req->r_request_started);
1610
1611                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1612                         ret = fill_readdir_cache(d_inode(parent), dn,
1613                                                  &cache_ctl, req);
1614                         if (ret < 0)
1615                                 err = ret;
1616                 }
1617 next_item:
1618                 if (dn)
1619                         dput(dn);
1620         }
1621 out:
1622         if (err == 0 && skipped == 0) {
1623                 req->r_did_prepopulate = true;
1624                 req->r_readdir_cache_idx = cache_ctl.index;
1625         }
1626         ceph_readdir_cache_release(&cache_ctl);
1627         if (snapdir) {
1628                 iput(snapdir);
1629                 dput(parent);
1630         }
1631         dout("readdir_prepopulate done\n");
1632         return err;
1633 }
1634
1635 int ceph_inode_set_size(struct inode *inode, loff_t size)
1636 {
1637         struct ceph_inode_info *ci = ceph_inode(inode);
1638         int ret = 0;
1639
1640         spin_lock(&ci->i_ceph_lock);
1641         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1642         i_size_write(inode, size);
1643         inode->i_blocks = calc_inode_blocks(size);
1644
1645         /* tell the MDS if we are approaching max_size */
1646         if ((size << 1) >= ci->i_max_size &&
1647             (ci->i_reported_size << 1) < ci->i_max_size)
1648                 ret = 1;
1649
1650         spin_unlock(&ci->i_ceph_lock);
1651         return ret;
1652 }
1653
1654 /*
1655  * Write back inode data in a worker thread.  (This can't be done
1656  * in the message handler context.)
1657  */
1658 void ceph_queue_writeback(struct inode *inode)
1659 {
1660         ihold(inode);
1661         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1662                        &ceph_inode(inode)->i_wb_work)) {
1663                 dout("ceph_queue_writeback %p\n", inode);
1664         } else {
1665                 dout("ceph_queue_writeback %p failed\n", inode);
1666                 iput(inode);
1667         }
1668 }
1669
1670 static void ceph_writeback_work(struct work_struct *work)
1671 {
1672         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1673                                                   i_wb_work);
1674         struct inode *inode = &ci->vfs_inode;
1675
1676         dout("writeback %p\n", inode);
1677         filemap_fdatawrite(&inode->i_data);
1678         iput(inode);
1679 }
1680
1681 /*
1682  * queue an async invalidation
1683  */
1684 void ceph_queue_invalidate(struct inode *inode)
1685 {
1686         ihold(inode);
1687         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1688                        &ceph_inode(inode)->i_pg_inv_work)) {
1689                 dout("ceph_queue_invalidate %p\n", inode);
1690         } else {
1691                 dout("ceph_queue_invalidate %p failed\n", inode);
1692                 iput(inode);
1693         }
1694 }
1695
1696 /*
1697  * Invalidate inode pages in a worker thread.  (This can't be done
1698  * in the message handler context.)
1699  */
1700 static void ceph_invalidate_work(struct work_struct *work)
1701 {
1702         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1703                                                   i_pg_inv_work);
1704         struct inode *inode = &ci->vfs_inode;
1705         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1706         u32 orig_gen;
1707         int check = 0;
1708
1709         mutex_lock(&ci->i_truncate_mutex);
1710
1711         if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1712                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1713                                     inode, ceph_ino(inode));
1714                 mapping_set_error(inode->i_mapping, -EIO);
1715                 truncate_pagecache(inode, 0);
1716                 mutex_unlock(&ci->i_truncate_mutex);
1717                 goto out;
1718         }
1719
1720         spin_lock(&ci->i_ceph_lock);
1721         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1722              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1723         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1724                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1725                         check = 1;
1726                 spin_unlock(&ci->i_ceph_lock);
1727                 mutex_unlock(&ci->i_truncate_mutex);
1728                 goto out;
1729         }
1730         orig_gen = ci->i_rdcache_gen;
1731         spin_unlock(&ci->i_ceph_lock);
1732
1733         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1734                 pr_err("invalidate_pages %p fails\n", inode);
1735         }
1736
1737         spin_lock(&ci->i_ceph_lock);
1738         if (orig_gen == ci->i_rdcache_gen &&
1739             orig_gen == ci->i_rdcache_revoking) {
1740                 dout("invalidate_pages %p gen %d successful\n", inode,
1741                      ci->i_rdcache_gen);
1742                 ci->i_rdcache_revoking--;
1743                 check = 1;
1744         } else {
1745                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1746                      inode, orig_gen, ci->i_rdcache_gen,
1747                      ci->i_rdcache_revoking);
1748                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1749                         check = 1;
1750         }
1751         spin_unlock(&ci->i_ceph_lock);
1752         mutex_unlock(&ci->i_truncate_mutex);
1753 out:
1754         if (check)
1755                 ceph_check_caps(ci, 0, NULL);
1756         iput(inode);
1757 }
1758
1759
1760 /*
1761  * called by trunc_wq;
1762  *
1763  * We also truncate in a separate thread as well.
1764  */
1765 static void ceph_vmtruncate_work(struct work_struct *work)
1766 {
1767         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1768                                                   i_vmtruncate_work);
1769         struct inode *inode = &ci->vfs_inode;
1770
1771         dout("vmtruncate_work %p\n", inode);
1772         __ceph_do_pending_vmtruncate(inode);
1773         iput(inode);
1774 }
1775
1776 /*
1777  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1778  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1779  */
1780 void ceph_queue_vmtruncate(struct inode *inode)
1781 {
1782         struct ceph_inode_info *ci = ceph_inode(inode);
1783
1784         ihold(inode);
1785
1786         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1787                        &ci->i_vmtruncate_work)) {
1788                 dout("ceph_queue_vmtruncate %p\n", inode);
1789         } else {
1790                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1791                      inode, ci->i_truncate_pending);
1792                 iput(inode);
1793         }
1794 }
1795
1796 /*
1797  * Make sure any pending truncation is applied before doing anything
1798  * that may depend on it.
1799  */
1800 void __ceph_do_pending_vmtruncate(struct inode *inode)
1801 {
1802         struct ceph_inode_info *ci = ceph_inode(inode);
1803         u64 to;
1804         int wrbuffer_refs, finish = 0;
1805
1806         mutex_lock(&ci->i_truncate_mutex);
1807 retry:
1808         spin_lock(&ci->i_ceph_lock);
1809         if (ci->i_truncate_pending == 0) {
1810                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1811                 spin_unlock(&ci->i_ceph_lock);
1812                 mutex_unlock(&ci->i_truncate_mutex);
1813                 return;
1814         }
1815
1816         /*
1817          * make sure any dirty snapped pages are flushed before we
1818          * possibly truncate them.. so write AND block!
1819          */
1820         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1821                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1822                      inode);
1823                 spin_unlock(&ci->i_ceph_lock);
1824                 filemap_write_and_wait_range(&inode->i_data, 0,
1825                                              inode->i_sb->s_maxbytes);
1826                 goto retry;
1827         }
1828
1829         /* there should be no reader or writer */
1830         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1831
1832         to = ci->i_truncate_size;
1833         wrbuffer_refs = ci->i_wrbuffer_ref;
1834         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1835              ci->i_truncate_pending, to);
1836         spin_unlock(&ci->i_ceph_lock);
1837
1838         truncate_pagecache(inode, to);
1839
1840         spin_lock(&ci->i_ceph_lock);
1841         if (to == ci->i_truncate_size) {
1842                 ci->i_truncate_pending = 0;
1843                 finish = 1;
1844         }
1845         spin_unlock(&ci->i_ceph_lock);
1846         if (!finish)
1847                 goto retry;
1848
1849         mutex_unlock(&ci->i_truncate_mutex);
1850
1851         if (wrbuffer_refs == 0)
1852                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1853
1854         wake_up_all(&ci->i_cap_wq);
1855 }
1856
1857 /*
1858  * symlinks
1859  */
1860 static const struct inode_operations ceph_symlink_iops = {
1861         .readlink = generic_readlink,
1862         .get_link = simple_get_link,
1863         .setattr = ceph_setattr,
1864         .getattr = ceph_getattr,
1865         .setxattr = generic_setxattr,
1866         .getxattr = generic_getxattr,
1867         .listxattr = ceph_listxattr,
1868         .removexattr = generic_removexattr,
1869 };
1870
1871 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1872 {
1873         struct ceph_inode_info *ci = ceph_inode(inode);
1874         const unsigned int ia_valid = attr->ia_valid;
1875         struct ceph_mds_request *req;
1876         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1877         struct ceph_cap_flush *prealloc_cf;
1878         int issued;
1879         int release = 0, dirtied = 0;
1880         int mask = 0;
1881         int err = 0;
1882         int inode_dirty_flags = 0;
1883         bool lock_snap_rwsem = false;
1884
1885         if (ceph_snap(inode) != CEPH_NOSNAP)
1886                 return -EROFS;
1887
1888         err = inode_change_ok(inode, attr);
1889         if (err != 0)
1890                 return err;
1891
1892         prealloc_cf = ceph_alloc_cap_flush();
1893         if (!prealloc_cf)
1894                 return -ENOMEM;
1895
1896         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1897                                        USE_AUTH_MDS);
1898         if (IS_ERR(req)) {
1899                 ceph_free_cap_flush(prealloc_cf);
1900                 return PTR_ERR(req);
1901         }
1902
1903         spin_lock(&ci->i_ceph_lock);
1904         issued = __ceph_caps_issued(ci, NULL);
1905
1906         if (!ci->i_head_snapc &&
1907             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1908                 lock_snap_rwsem = true;
1909                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1910                         spin_unlock(&ci->i_ceph_lock);
1911                         down_read(&mdsc->snap_rwsem);
1912                         spin_lock(&ci->i_ceph_lock);
1913                         issued = __ceph_caps_issued(ci, NULL);
1914                 }
1915         }
1916
1917         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1918
1919         if (ia_valid & ATTR_UID) {
1920                 dout("setattr %p uid %d -> %d\n", inode,
1921                      from_kuid(&init_user_ns, inode->i_uid),
1922                      from_kuid(&init_user_ns, attr->ia_uid));
1923                 if (issued & CEPH_CAP_AUTH_EXCL) {
1924                         inode->i_uid = attr->ia_uid;
1925                         dirtied |= CEPH_CAP_AUTH_EXCL;
1926                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1927                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1928                         req->r_args.setattr.uid = cpu_to_le32(
1929                                 from_kuid(&init_user_ns, attr->ia_uid));
1930                         mask |= CEPH_SETATTR_UID;
1931                         release |= CEPH_CAP_AUTH_SHARED;
1932                 }
1933         }
1934         if (ia_valid & ATTR_GID) {
1935                 dout("setattr %p gid %d -> %d\n", inode,
1936                      from_kgid(&init_user_ns, inode->i_gid),
1937                      from_kgid(&init_user_ns, attr->ia_gid));
1938                 if (issued & CEPH_CAP_AUTH_EXCL) {
1939                         inode->i_gid = attr->ia_gid;
1940                         dirtied |= CEPH_CAP_AUTH_EXCL;
1941                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1942                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1943                         req->r_args.setattr.gid = cpu_to_le32(
1944                                 from_kgid(&init_user_ns, attr->ia_gid));
1945                         mask |= CEPH_SETATTR_GID;
1946                         release |= CEPH_CAP_AUTH_SHARED;
1947                 }
1948         }
1949         if (ia_valid & ATTR_MODE) {
1950                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1951                      attr->ia_mode);
1952                 if (issued & CEPH_CAP_AUTH_EXCL) {
1953                         inode->i_mode = attr->ia_mode;
1954                         dirtied |= CEPH_CAP_AUTH_EXCL;
1955                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1956                            attr->ia_mode != inode->i_mode) {
1957                         inode->i_mode = attr->ia_mode;
1958                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1959                         mask |= CEPH_SETATTR_MODE;
1960                         release |= CEPH_CAP_AUTH_SHARED;
1961                 }
1962         }
1963
1964         if (ia_valid & ATTR_ATIME) {
1965                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1966                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1967                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1968                 if (issued & CEPH_CAP_FILE_EXCL) {
1969                         ci->i_time_warp_seq++;
1970                         inode->i_atime = attr->ia_atime;
1971                         dirtied |= CEPH_CAP_FILE_EXCL;
1972                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1973                            timespec_compare(&inode->i_atime,
1974                                             &attr->ia_atime) < 0) {
1975                         inode->i_atime = attr->ia_atime;
1976                         dirtied |= CEPH_CAP_FILE_WR;
1977                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1978                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1979                         ceph_encode_timespec(&req->r_args.setattr.atime,
1980                                              &attr->ia_atime);
1981                         mask |= CEPH_SETATTR_ATIME;
1982                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1983                                 CEPH_CAP_FILE_WR;
1984                 }
1985         }
1986         if (ia_valid & ATTR_MTIME) {
1987                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1988                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1989                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1990                 if (issued & CEPH_CAP_FILE_EXCL) {
1991                         ci->i_time_warp_seq++;
1992                         inode->i_mtime = attr->ia_mtime;
1993                         dirtied |= CEPH_CAP_FILE_EXCL;
1994                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1995                            timespec_compare(&inode->i_mtime,
1996                                             &attr->ia_mtime) < 0) {
1997                         inode->i_mtime = attr->ia_mtime;
1998                         dirtied |= CEPH_CAP_FILE_WR;
1999                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2000                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2001                         ceph_encode_timespec(&req->r_args.setattr.mtime,
2002                                              &attr->ia_mtime);
2003                         mask |= CEPH_SETATTR_MTIME;
2004                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2005                                 CEPH_CAP_FILE_WR;
2006                 }
2007         }
2008         if (ia_valid & ATTR_SIZE) {
2009                 dout("setattr %p size %lld -> %lld\n", inode,
2010                      inode->i_size, attr->ia_size);
2011                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2012                     attr->ia_size > inode->i_size) {
2013                         i_size_write(inode, attr->ia_size);
2014                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2015                         inode->i_ctime = attr->ia_ctime;
2016                         ci->i_reported_size = attr->ia_size;
2017                         dirtied |= CEPH_CAP_FILE_EXCL;
2018                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2019                            attr->ia_size != inode->i_size) {
2020                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2021                         req->r_args.setattr.old_size =
2022                                 cpu_to_le64(inode->i_size);
2023                         mask |= CEPH_SETATTR_SIZE;
2024                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2025                                 CEPH_CAP_FILE_WR;
2026                 }
2027         }
2028
2029         /* these do nothing */
2030         if (ia_valid & ATTR_CTIME) {
2031                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2032                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2033                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2034                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2035                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2036                      only ? "ctime only" : "ignored");
2037                 inode->i_ctime = attr->ia_ctime;
2038                 if (only) {
2039                         /*
2040                          * if kernel wants to dirty ctime but nothing else,
2041                          * we need to choose a cap to dirty under, or do
2042                          * a almost-no-op setattr
2043                          */
2044                         if (issued & CEPH_CAP_AUTH_EXCL)
2045                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2046                         else if (issued & CEPH_CAP_FILE_EXCL)
2047                                 dirtied |= CEPH_CAP_FILE_EXCL;
2048                         else if (issued & CEPH_CAP_XATTR_EXCL)
2049                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2050                         else
2051                                 mask |= CEPH_SETATTR_CTIME;
2052                 }
2053         }
2054         if (ia_valid & ATTR_FILE)
2055                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2056
2057         if (dirtied) {
2058                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2059                                                            &prealloc_cf);
2060                 inode->i_ctime = current_fs_time(inode->i_sb);
2061         }
2062
2063         release &= issued;
2064         spin_unlock(&ci->i_ceph_lock);
2065         if (lock_snap_rwsem)
2066                 up_read(&mdsc->snap_rwsem);
2067
2068         if (inode_dirty_flags)
2069                 __mark_inode_dirty(inode, inode_dirty_flags);
2070
2071         if (ia_valid & ATTR_MODE) {
2072                 err = posix_acl_chmod(inode, attr->ia_mode);
2073                 if (err)
2074                         goto out_put;
2075         }
2076
2077         if (mask) {
2078                 req->r_inode = inode;
2079                 ihold(inode);
2080                 req->r_inode_drop = release;
2081                 req->r_args.setattr.mask = cpu_to_le32(mask);
2082                 req->r_num_caps = 1;
2083                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2084         }
2085         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2086              ceph_cap_string(dirtied), mask);
2087
2088         ceph_mdsc_put_request(req);
2089         if (mask & CEPH_SETATTR_SIZE)
2090                 __ceph_do_pending_vmtruncate(inode);
2091         ceph_free_cap_flush(prealloc_cf);
2092         return err;
2093 out_put:
2094         ceph_mdsc_put_request(req);
2095         ceph_free_cap_flush(prealloc_cf);
2096         return err;
2097 }
2098
2099 /*
2100  * setattr
2101  */
2102 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2103 {
2104         return __ceph_setattr(d_inode(dentry), attr);
2105 }
2106
2107 /*
2108  * Verify that we have a lease on the given mask.  If not,
2109  * do a getattr against an mds.
2110  */
2111 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2112                       int mask, bool force)
2113 {
2114         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2115         struct ceph_mds_client *mdsc = fsc->mdsc;
2116         struct ceph_mds_request *req;
2117         int err;
2118
2119         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2120                 dout("do_getattr inode %p SNAPDIR\n", inode);
2121                 return 0;
2122         }
2123
2124         dout("do_getattr inode %p mask %s mode 0%o\n",
2125              inode, ceph_cap_string(mask), inode->i_mode);
2126         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2127                 return 0;
2128
2129         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2130         if (IS_ERR(req))
2131                 return PTR_ERR(req);
2132         req->r_inode = inode;
2133         ihold(inode);
2134         req->r_num_caps = 1;
2135         req->r_args.getattr.mask = cpu_to_le32(mask);
2136         req->r_locked_page = locked_page;
2137         err = ceph_mdsc_do_request(mdsc, NULL, req);
2138         if (locked_page && err == 0) {
2139                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2140                 if (inline_version == 0) {
2141                         /* the reply is supposed to contain inline data */
2142                         err = -EINVAL;
2143                 } else if (inline_version == CEPH_INLINE_NONE) {
2144                         err = -ENODATA;
2145                 } else {
2146                         err = req->r_reply_info.targeti.inline_len;
2147                 }
2148         }
2149         ceph_mdsc_put_request(req);
2150         dout("do_getattr result=%d\n", err);
2151         return err;
2152 }
2153
2154
2155 /*
2156  * Check inode permissions.  We verify we have a valid value for
2157  * the AUTH cap, then call the generic handler.
2158  */
2159 int ceph_permission(struct inode *inode, int mask)
2160 {
2161         int err;
2162
2163         if (mask & MAY_NOT_BLOCK)
2164                 return -ECHILD;
2165
2166         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2167
2168         if (!err)
2169                 err = generic_permission(inode, mask);
2170         return err;
2171 }
2172
2173 /*
2174  * Get all attributes.  Hopefully somedata we'll have a statlite()
2175  * and can limit the fields we require to be accurate.
2176  */
2177 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2178                  struct kstat *stat)
2179 {
2180         struct inode *inode = d_inode(dentry);
2181         struct ceph_inode_info *ci = ceph_inode(inode);
2182         int err;
2183
2184         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2185         if (!err) {
2186                 generic_fillattr(inode, stat);
2187                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2188                 if (ceph_snap(inode) != CEPH_NOSNAP)
2189                         stat->dev = ceph_snap(inode);
2190                 else
2191                         stat->dev = 0;
2192                 if (S_ISDIR(inode->i_mode)) {
2193                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2194                                                 RBYTES))
2195                                 stat->size = ci->i_rbytes;
2196                         else
2197                                 stat->size = ci->i_files + ci->i_subdirs;
2198                         stat->blocks = 0;
2199                         stat->blksize = 65536;
2200                 }
2201         }
2202         return err;
2203 }