fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16
  17 #include "super.h"
  18 #include "mds_client.h"
  19 #include "cache.h"
  20 #include <linux/ceph/decode.h>
  21
  22 /*
  23  * Ceph inode operations
  24  *
  25  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  26  * setattr, etc.), xattr helpers, and helpers for assimilating
  27  * metadata returned by the MDS into our cache.
  28  *
  29  * Also define helpers for doing asynchronous writeback, invalidation,
  30  * and truncation for the benefit of those who can't afford to block
  31  * (typically because they are in the message handler path).
  32  */
  33
  34 static const struct inode_operations ceph_symlink_iops;
  35
  36 static void ceph_invalidate_work(struct work_struct *work);
  37 static void ceph_writeback_work(struct work_struct *work);
  38 static void ceph_vmtruncate_work(struct work_struct *work);
  39
  40 /*
  41  * find or create an inode, given the ceph ino number
  42  */
  43 static int ceph_set_ino_cb(struct inode *inode, void *data)
  44 {
  45         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  46         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  47         return 0;
  48 }
  49
  50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  51 {
  52         struct inode *inode;
  53         ino_t t = ceph_vino_to_ino(vino);
  54
  55         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  56         if (!inode)
  57                 return ERR_PTR(-ENOMEM);
  58         if (inode->i_state & I_NEW) {
  59                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  60                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  61                 unlock_new_inode(inode);
  62         }
  63
  64         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  65              vino.snap, inode);
  66         return inode;
  67 }
  68
  69 /*
  70  * get/constuct snapdir inode for a given directory
  71  */
  72 struct inode *ceph_get_snapdir(struct inode *parent)
  73 {
  74         struct ceph_vino vino = {
  75                 .ino = ceph_ino(parent),
  76                 .snap = CEPH_SNAPDIR,
  77         };
  78         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  79         struct ceph_inode_info *ci = ceph_inode(inode);
  80
  81         BUG_ON(!S_ISDIR(parent->i_mode));
  82         if (IS_ERR(inode))
  83                 return inode;
  84         inode->i_mode = parent->i_mode;
  85         inode->i_uid = parent->i_uid;
  86         inode->i_gid = parent->i_gid;
  87         inode->i_op = &ceph_snapdir_iops;
  88         inode->i_fop = &ceph_snapdir_fops;
  89         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  90         ci->i_rbytes = 0;
  91         return inode;
  92 }
  93
  94 const struct inode_operations ceph_file_iops = {
  95         .permission = ceph_permission,
  96         .setattr = ceph_setattr,
  97         .getattr = ceph_getattr,
  98         .listxattr = ceph_listxattr,
  99         .get_acl = ceph_get_acl,
 100         .set_acl = ceph_set_acl,
 101 };
 102
 103
 104 /*
 105  * We use a 'frag tree' to keep track of the MDS's directory fragments
 106  * for a given inode (usually there is just a single fragment).  We
 107  * need to know when a child frag is delegated to a new MDS, or when
 108  * it is flagged as replicated, so we can direct our requests
 109  * accordingly.
 110  */
 111
 112 /*
 113  * find/create a frag in the tree
 114  */
 115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 116                                                     u32 f)
 117 {
 118         struct rb_node **p;
 119         struct rb_node *parent = NULL;
 120         struct ceph_inode_frag *frag;
 121         int c;
 122
 123         p = &ci->i_fragtree.rb_node;
 124         while (*p) {
 125                 parent = *p;
 126                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 127                 c = ceph_frag_compare(f, frag->frag);
 128                 if (c < 0)
 129                         p = &(*p)->rb_left;
 130                 else if (c > 0)
 131                         p = &(*p)->rb_right;
 132                 else
 133                         return frag;
 134         }
 135
 136         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 137         if (!frag)
 138                 return ERR_PTR(-ENOMEM);
 139
 140         frag->frag = f;
 141         frag->split_by = 0;
 142         frag->mds = -1;
 143         frag->ndist = 0;
 144
 145         rb_link_node(&frag->node, parent, p);
 146         rb_insert_color(&frag->node, &ci->i_fragtree);
 147
 148         dout("get_or_create_frag added %llx.%llx frag %x\n",
 149              ceph_vinop(&ci->vfs_inode), f);
 150         return frag;
 151 }
 152
 153 /*
 154  * find a specific frag @f
 155  */
 156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 157 {
 158         struct rb_node *n = ci->i_fragtree.rb_node;
 159
 160         while (n) {
 161                 struct ceph_inode_frag *frag =
 162                         rb_entry(n, struct ceph_inode_frag, node);
 163                 int c = ceph_frag_compare(f, frag->frag);
 164                 if (c < 0)
 165                         n = n->rb_left;
 166                 else if (c > 0)
 167                         n = n->rb_right;
 168                 else
 169                         return frag;
 170         }
 171         return NULL;
 172 }
 173
 174 /*
 175  * Choose frag containing the given value @v.  If @pfrag is
 176  * specified, copy the frag delegation info to the caller if
 177  * it is present.
 178  */
 179 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 180                               struct ceph_inode_frag *pfrag, int *found)
 181 {
 182         u32 t = ceph_frag_make(0, 0);
 183         struct ceph_inode_frag *frag;
 184         unsigned nway, i;
 185         u32 n;
 186
 187         if (found)
 188                 *found = 0;
 189
 190         while (1) {
 191                 WARN_ON(!ceph_frag_contains_value(t, v));
 192                 frag = __ceph_find_frag(ci, t);
 193                 if (!frag)
 194                         break; /* t is a leaf */
 195                 if (frag->split_by == 0) {
 196                         if (pfrag)
 197                                 memcpy(pfrag, frag, sizeof(*pfrag));
 198                         if (found)
 199                                 *found = 1;
 200                         break;
 201                 }
 202
 203                 /* choose child */
 204                 nway = 1 << frag->split_by;
 205                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 206                      frag->split_by, nway);
 207                 for (i = 0; i < nway; i++) {
 208                         n = ceph_frag_make_child(t, frag->split_by, i);
 209                         if (ceph_frag_contains_value(n, v)) {
 210                                 t = n;
 211                                 break;
 212                         }
 213                 }
 214                 BUG_ON(i == nway);
 215         }
 216         dout("choose_frag(%x) = %x\n", v, t);
 217
 218         return t;
 219 }
 220
 221 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 222                      struct ceph_inode_frag *pfrag, int *found)
 223 {
 224         u32 ret;
 225         mutex_lock(&ci->i_fragtree_mutex);
 226         ret = __ceph_choose_frag(ci, v, pfrag, found);
 227         mutex_unlock(&ci->i_fragtree_mutex);
 228         return ret;
 229 }
 230
 231 /*
 232  * Process dirfrag (delegation) info from the mds.  Include leaf
 233  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 234  * branches/splits are included in i_fragtree)
 235  */
 236 static int ceph_fill_dirfrag(struct inode *inode,
 237                              struct ceph_mds_reply_dirfrag *dirinfo)
 238 {
 239         struct ceph_inode_info *ci = ceph_inode(inode);
 240         struct ceph_inode_frag *frag;
 241         u32 id = le32_to_cpu(dirinfo->frag);
 242         int mds = le32_to_cpu(dirinfo->auth);
 243         int ndist = le32_to_cpu(dirinfo->ndist);
 244         int diri_auth = -1;
 245         int i;
 246         int err = 0;
 247
 248         spin_lock(&ci->i_ceph_lock);
 249         if (ci->i_auth_cap)
 250                 diri_auth = ci->i_auth_cap->mds;
 251         spin_unlock(&ci->i_ceph_lock);
 252
 253         if (mds == -1) /* CDIR_AUTH_PARENT */
 254                 mds = diri_auth;
 255
 256         mutex_lock(&ci->i_fragtree_mutex);
 257         if (ndist == 0 && mds == diri_auth) {
 258                 /* no delegation info needed. */
 259                 frag = __ceph_find_frag(ci, id);
 260                 if (!frag)
 261                         goto out;
 262                 if (frag->split_by == 0) {
 263                         /* tree leaf, remove */
 264                         dout("fill_dirfrag removed %llx.%llx frag %x"
 265                              " (no ref)\n", ceph_vinop(inode), id);
 266                         rb_erase(&frag->node, &ci->i_fragtree);
 267                         kfree(frag);
 268                 } else {
 269                         /* tree branch, keep and clear */
 270                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 271                              " referral\n", ceph_vinop(inode), id);
 272                         frag->mds = -1;
 273                         frag->ndist = 0;
 274                 }
 275                 goto out;
 276         }
 277
 278
 279         /* find/add this frag to store mds delegation info */
 280         frag = __get_or_create_frag(ci, id);
 281         if (IS_ERR(frag)) {
 282                 /* this is not the end of the world; we can continue
 283                    with bad/inaccurate delegation info */
 284                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 285                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 286                 err = -ENOMEM;
 287                 goto out;
 288         }
 289
 290         frag->mds = mds;
 291         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 292         for (i = 0; i < frag->ndist; i++)
 293                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 294         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 295              ceph_vinop(inode), frag->frag, frag->ndist);
 296
 297 out:
 298         mutex_unlock(&ci->i_fragtree_mutex);
 299         return err;
 300 }
 301
 302 static int frag_tree_split_cmp(const void *l, const void *r)
 303 {
 304         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 305         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 306         return ceph_frag_compare(le32_to_cpu(ls->frag),
 307                                  le32_to_cpu(rs->frag));
 308 }
 309
 310 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 311 {
 312         if (!frag)
 313                 return f == ceph_frag_make(0, 0);
 314         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 315                 return false;
 316         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 317 }
 318
 319 static int ceph_fill_fragtree(struct inode *inode,
 320                               struct ceph_frag_tree_head *fragtree,
 321                               struct ceph_mds_reply_dirfrag *dirinfo)
 322 {
 323         struct ceph_inode_info *ci = ceph_inode(inode);
 324         struct ceph_inode_frag *frag, *prev_frag = NULL;
 325         struct rb_node *rb_node;
 326         unsigned i, split_by, nsplits;
 327         u32 id;
 328         bool update = false;
 329
 330         mutex_lock(&ci->i_fragtree_mutex);
 331         nsplits = le32_to_cpu(fragtree->nsplits);
 332         if (nsplits != ci->i_fragtree_nsplits) {
 333                 update = true;
 334         } else if (nsplits) {
 335                 i = prandom_u32() % nsplits;
 336                 id = le32_to_cpu(fragtree->splits[i].frag);
 337                 if (!__ceph_find_frag(ci, id))
 338                         update = true;
 339         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 340                 rb_node = rb_first(&ci->i_fragtree);
 341                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 342                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 343                         update = true;
 344         }
 345         if (!update && dirinfo) {
 346                 id = le32_to_cpu(dirinfo->frag);
 347                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 348                         update = true;
 349         }
 350         if (!update)
 351                 goto out_unlock;
 352
 353         if (nsplits > 1) {
 354                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 355                      frag_tree_split_cmp, NULL);
 356         }
 357
 358         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 359         rb_node = rb_first(&ci->i_fragtree);
 360         for (i = 0; i < nsplits; i++) {
 361                 id = le32_to_cpu(fragtree->splits[i].frag);
 362                 split_by = le32_to_cpu(fragtree->splits[i].by);
 363                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 364                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 365                                "frag %x split by %d\n", ceph_vinop(inode),
 366                                i, nsplits, id, split_by);
 367                         continue;
 368                 }
 369                 frag = NULL;
 370                 while (rb_node) {
 371                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 372                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 373                                 if (frag->frag != id)
 374                                         frag = NULL;
 375                                 else
 376                                         rb_node = rb_next(rb_node);
 377                                 break;
 378                         }
 379                         rb_node = rb_next(rb_node);
 380                         /* delete stale split/leaf node */
 381                         if (frag->split_by > 0 ||
 382                             !is_frag_child(frag->frag, prev_frag)) {
 383                                 rb_erase(&frag->node, &ci->i_fragtree);
 384                                 if (frag->split_by > 0)
 385                                         ci->i_fragtree_nsplits--;
 386                                 kfree(frag);
 387                         }
 388                         frag = NULL;
 389                 }
 390                 if (!frag) {
 391                         frag = __get_or_create_frag(ci, id);
 392                         if (IS_ERR(frag))
 393                                 continue;
 394                 }
 395                 if (frag->split_by == 0)
 396                         ci->i_fragtree_nsplits++;
 397                 frag->split_by = split_by;
 398                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 399                 prev_frag = frag;
 400         }
 401         while (rb_node) {
 402                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 403                 rb_node = rb_next(rb_node);
 404                 /* delete stale split/leaf node */
 405                 if (frag->split_by > 0 ||
 406                     !is_frag_child(frag->frag, prev_frag)) {
 407                         rb_erase(&frag->node, &ci->i_fragtree);
 408                         if (frag->split_by > 0)
 409                                 ci->i_fragtree_nsplits--;
 410                         kfree(frag);
 411                 }
 412         }
 413 out_unlock:
 414         mutex_unlock(&ci->i_fragtree_mutex);
 415         return 0;
 416 }
 417
 418 /*
 419  * initialize a newly allocated inode.
 420  */
 421 struct inode *ceph_alloc_inode(struct super_block *sb)
 422 {
 423         struct ceph_inode_info *ci;
 424         int i;
 425
 426         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 427         if (!ci)
 428                 return NULL;
 429
 430         dout("alloc_inode %p\n", &ci->vfs_inode);
 431
 432         spin_lock_init(&ci->i_ceph_lock);
 433
 434         ci->i_version = 0;
 435         ci->i_inline_version = 0;
 436         ci->i_time_warp_seq = 0;
 437         ci->i_ceph_flags = 0;
 438         atomic64_set(&ci->i_ordered_count, 1);
 439         atomic64_set(&ci->i_release_count, 1);
 440         atomic64_set(&ci->i_complete_seq[0], 0);
 441         atomic64_set(&ci->i_complete_seq[1], 0);
 442         ci->i_symlink = NULL;
 443
 444         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 445         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 446
 447         ci->i_fragtree = RB_ROOT;
 448         mutex_init(&ci->i_fragtree_mutex);
 449
 450         ci->i_xattrs.blob = NULL;
 451         ci->i_xattrs.prealloc_blob = NULL;
 452         ci->i_xattrs.dirty = false;
 453         ci->i_xattrs.index = RB_ROOT;
 454         ci->i_xattrs.count = 0;
 455         ci->i_xattrs.names_size = 0;
 456         ci->i_xattrs.vals_size = 0;
 457         ci->i_xattrs.version = 0;
 458         ci->i_xattrs.index_version = 0;
 459
 460         ci->i_caps = RB_ROOT;
 461         ci->i_auth_cap = NULL;
 462         ci->i_dirty_caps = 0;
 463         ci->i_flushing_caps = 0;
 464         INIT_LIST_HEAD(&ci->i_dirty_item);
 465         INIT_LIST_HEAD(&ci->i_flushing_item);
 466         ci->i_prealloc_cap_flush = NULL;
 467         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 468         init_waitqueue_head(&ci->i_cap_wq);
 469         ci->i_hold_caps_min = 0;
 470         ci->i_hold_caps_max = 0;
 471         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 472         INIT_LIST_HEAD(&ci->i_cap_snaps);
 473         ci->i_head_snapc = NULL;
 474         ci->i_snap_caps = 0;
 475
 476         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 477                 ci->i_nr_by_mode[i] = 0;
 478
 479         mutex_init(&ci->i_truncate_mutex);
 480         ci->i_truncate_seq = 0;
 481         ci->i_truncate_size = 0;
 482         ci->i_truncate_pending = 0;
 483
 484         ci->i_max_size = 0;
 485         ci->i_reported_size = 0;
 486         ci->i_wanted_max_size = 0;
 487         ci->i_requested_max_size = 0;
 488
 489         ci->i_pin_ref = 0;
 490         ci->i_rd_ref = 0;
 491         ci->i_rdcache_ref = 0;
 492         ci->i_wr_ref = 0;
 493         ci->i_wb_ref = 0;
 494         ci->i_wrbuffer_ref = 0;
 495         ci->i_wrbuffer_ref_head = 0;
 496         ci->i_shared_gen = 0;
 497         ci->i_rdcache_gen = 0;
 498         ci->i_rdcache_revoking = 0;
 499
 500         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 501         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 502         spin_lock_init(&ci->i_unsafe_lock);
 503
 504         ci->i_snap_realm = NULL;
 505         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 506         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 507
 508         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 509         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 510
 511         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 512
 513         ceph_fscache_inode_init(ci);
 514
 515         return &ci->vfs_inode;
 516 }
 517
 518 static void ceph_i_callback(struct rcu_head *head)
 519 {
 520         struct inode *inode = container_of(head, struct inode, i_rcu);
 521         struct ceph_inode_info *ci = ceph_inode(inode);
 522
 523         kmem_cache_free(ceph_inode_cachep, ci);
 524 }
 525
 526 void ceph_destroy_inode(struct inode *inode)
 527 {
 528         struct ceph_inode_info *ci = ceph_inode(inode);
 529         struct ceph_inode_frag *frag;
 530         struct rb_node *n;
 531
 532         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 533
 534         ceph_fscache_unregister_inode_cookie(ci);
 535
 536         ceph_queue_caps_release(inode);
 537
 538         /*
 539          * we may still have a snap_realm reference if there are stray
 540          * caps in i_snap_caps.
 541          */
 542         if (ci->i_snap_realm) {
 543                 struct ceph_mds_client *mdsc =
 544                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 545                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 546
 547                 dout(" dropping residual ref to snap realm %p\n", realm);
 548                 spin_lock(&realm->inodes_with_caps_lock);
 549                 list_del_init(&ci->i_snap_realm_item);
 550                 spin_unlock(&realm->inodes_with_caps_lock);
 551                 ceph_put_snap_realm(mdsc, realm);
 552         }
 553
 554         kfree(ci->i_symlink);
 555         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 556                 frag = rb_entry(n, struct ceph_inode_frag, node);
 557                 rb_erase(n, &ci->i_fragtree);
 558                 kfree(frag);
 559         }
 560         ci->i_fragtree_nsplits = 0;
 561
 562         __ceph_destroy_xattrs(ci);
 563         if (ci->i_xattrs.blob)
 564                 ceph_buffer_put(ci->i_xattrs.blob);
 565         if (ci->i_xattrs.prealloc_blob)
 566                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 567
 568         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 569
 570         call_rcu(&inode->i_rcu, ceph_i_callback);
 571 }
 572
 573 int ceph_drop_inode(struct inode *inode)
 574 {
 575         /*
 576          * Positve dentry and corresponding inode are always accompanied
 577          * in MDS reply. So no need to keep inode in the cache after
 578          * dropping all its aliases.
 579          */
 580         return 1;
 581 }
 582
 583 static inline blkcnt_t calc_inode_blocks(u64 size)
 584 {
 585         return (size + (1<<9) - 1) >> 9;
 586 }
 587
 588 /*
 589  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 590  * careful because either the client or MDS may have more up to date
 591  * info, depending on which capabilities are held, and whether
 592  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 593  * and size are monotonically increasing, except when utimes() or
 594  * truncate() increments the corresponding _seq values.)
 595  */
 596 int ceph_fill_file_size(struct inode *inode, int issued,
 597                         u32 truncate_seq, u64 truncate_size, u64 size)
 598 {
 599         struct ceph_inode_info *ci = ceph_inode(inode);
 600         int queue_trunc = 0;
 601
 602         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 603             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 604                 dout("size %lld -> %llu\n", inode->i_size, size);
 605                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 606                         pr_err("fill_file_size non-zero size for directory\n");
 607                         size = 0;
 608                 }
 609                 i_size_write(inode, size);
 610                 inode->i_blocks = calc_inode_blocks(size);
 611                 ci->i_reported_size = size;
 612                 if (truncate_seq != ci->i_truncate_seq) {
 613                         dout("truncate_seq %u -> %u\n",
 614                              ci->i_truncate_seq, truncate_seq);
 615                         ci->i_truncate_seq = truncate_seq;
 616
 617                         /* the MDS should have revoked these caps */
 618                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 619                                                CEPH_CAP_FILE_RD |
 620                                                CEPH_CAP_FILE_WR |
 621                                                CEPH_CAP_FILE_LAZYIO));
 622                         /*
 623                          * If we hold relevant caps, or in the case where we're
 624                          * not the only client referencing this file and we
 625                          * don't hold those caps, then we need to check whether
 626                          * the file is either opened or mmaped
 627                          */
 628                         if ((issued & (CEPH_CAP_FILE_CACHE|
 629                                        CEPH_CAP_FILE_BUFFER)) ||
 630                             mapping_mapped(inode->i_mapping) ||
 631                             __ceph_caps_file_wanted(ci)) {
 632                                 ci->i_truncate_pending++;
 633                                 queue_trunc = 1;
 634                         }
 635                 }
 636         }
 637         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 638             ci->i_truncate_size != truncate_size) {
 639                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 640                      truncate_size);
 641                 ci->i_truncate_size = truncate_size;
 642         }
 643
 644         if (queue_trunc)
 645                 ceph_fscache_invalidate(inode);
 646
 647         return queue_trunc;
 648 }
 649
 650 void ceph_fill_file_time(struct inode *inode, int issued,
 651                          u64 time_warp_seq, struct timespec *ctime,
 652                          struct timespec *mtime, struct timespec *atime)
 653 {
 654         struct ceph_inode_info *ci = ceph_inode(inode);
 655         int warn = 0;
 656
 657         if (issued & (CEPH_CAP_FILE_EXCL|
 658                       CEPH_CAP_FILE_WR|
 659                       CEPH_CAP_FILE_BUFFER|
 660                       CEPH_CAP_AUTH_EXCL|
 661                       CEPH_CAP_XATTR_EXCL)) {
 662                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 663                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 664                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 665                              ctime->tv_sec, ctime->tv_nsec);
 666                         inode->i_ctime = *ctime;
 667                 }
 668                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 669                         /* the MDS did a utimes() */
 670                         dout("mtime %ld.%09ld -> %ld.%09ld "
 671                              "tw %d -> %d\n",
 672                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 673                              mtime->tv_sec, mtime->tv_nsec,
 674                              ci->i_time_warp_seq, (int)time_warp_seq);
 675
 676                         inode->i_mtime = *mtime;
 677                         inode->i_atime = *atime;
 678                         ci->i_time_warp_seq = time_warp_seq;
 679                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 680                         /* nobody did utimes(); take the max */
 681                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 682                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 683                                      inode->i_mtime.tv_sec,
 684                                      inode->i_mtime.tv_nsec,
 685                                      mtime->tv_sec, mtime->tv_nsec);
 686                                 inode->i_mtime = *mtime;
 687                         }
 688                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 689                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 690                                      inode->i_atime.tv_sec,
 691                                      inode->i_atime.tv_nsec,
 692                                      atime->tv_sec, atime->tv_nsec);
 693                                 inode->i_atime = *atime;
 694                         }
 695                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 696                         /* we did a utimes(); ignore mds values */
 697                 } else {
 698                         warn = 1;
 699                 }
 700         } else {
 701                 /* we have no write|excl caps; whatever the MDS says is true */
 702                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 703                         inode->i_ctime = *ctime;
 704                         inode->i_mtime = *mtime;
 705                         inode->i_atime = *atime;
 706                         ci->i_time_warp_seq = time_warp_seq;
 707                 } else {
 708                         warn = 1;
 709                 }
 710         }
 711         if (warn) /* time_warp_seq shouldn't go backwards */
 712                 dout("%p mds time_warp_seq %llu < %u\n",
 713                      inode, time_warp_seq, ci->i_time_warp_seq);
 714 }
 715
 716 /*
 717  * Populate an inode based on info from mds.  May be called on new or
 718  * existing inodes.
 719  */
 720 static int fill_inode(struct inode *inode, struct page *locked_page,
 721                       struct ceph_mds_reply_info_in *iinfo,
 722                       struct ceph_mds_reply_dirfrag *dirinfo,
 723                       struct ceph_mds_session *session,
 724                       unsigned long ttl_from, int cap_fmode,
 725                       struct ceph_cap_reservation *caps_reservation)
 726 {
 727         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 728         struct ceph_mds_reply_inode *info = iinfo->in;
 729         struct ceph_inode_info *ci = ceph_inode(inode);
 730         int issued = 0, implemented, new_issued;
 731         struct timespec mtime, atime, ctime;
 732         struct ceph_buffer *xattr_blob = NULL;
 733         struct ceph_string *pool_ns = NULL;
 734         struct ceph_cap *new_cap = NULL;
 735         int err = 0;
 736         bool wake = false;
 737         bool queue_trunc = false;
 738         bool new_version = false;
 739         bool fill_inline = false;
 740
 741         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 742              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 743              ci->i_version);
 744
 745         /* prealloc new cap struct */
 746         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
 747                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 748
 749         /*
 750          * prealloc xattr data, if it looks like we'll need it.  only
 751          * if len > 4 (meaning there are actually xattrs; the first 4
 752          * bytes are the xattr count).
 753          */
 754         if (iinfo->xattr_len > 4) {
 755                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 756                 if (!xattr_blob)
 757                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 758                                iinfo->xattr_len);
 759         }
 760
 761         if (iinfo->pool_ns_len > 0)
 762                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 763                                                      iinfo->pool_ns_len);
 764
 765         spin_lock(&ci->i_ceph_lock);
 766
 767         /*
 768          * provided version will be odd if inode value is projected,
 769          * even if stable.  skip the update if we have newer stable
 770          * info (ours>=theirs, e.g. due to racing mds replies), unless
 771          * we are getting projected (unstable) info (in which case the
 772          * version is odd, and we want ours>theirs).
 773          *   us   them
 774          *   2    2     skip
 775          *   3    2     skip
 776          *   3    3     update
 777          */
 778         if (ci->i_version == 0 ||
 779             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 780              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 781                 new_version = true;
 782
 783         issued = __ceph_caps_issued(ci, &implemented);
 784         issued |= implemented | __ceph_caps_dirty(ci);
 785         new_issued = ~issued & le32_to_cpu(info->cap.caps);
 786
 787         /* update inode */
 788         ci->i_version = le64_to_cpu(info->version);
 789         inode->i_version++;
 790         inode->i_rdev = le32_to_cpu(info->rdev);
 791         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 792
 793         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 794             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 795                 inode->i_mode = le32_to_cpu(info->mode);
 796                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 797                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 798                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 799                      from_kuid(&init_user_ns, inode->i_uid),
 800                      from_kgid(&init_user_ns, inode->i_gid));
 801         }
 802
 803         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 804             (issued & CEPH_CAP_LINK_EXCL) == 0)
 805                 set_nlink(inode, le32_to_cpu(info->nlink));
 806
 807         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 808                 /* be careful with mtime, atime, size */
 809                 ceph_decode_timespec(&atime, &info->atime);
 810                 ceph_decode_timespec(&mtime, &info->mtime);
 811                 ceph_decode_timespec(&ctime, &info->ctime);
 812                 ceph_fill_file_time(inode, issued,
 813                                 le32_to_cpu(info->time_warp_seq),
 814                                 &ctime, &mtime, &atime);
 815         }
 816
 817         if (new_version ||
 818             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 819                 s64 old_pool = ci->i_layout.pool_id;
 820                 struct ceph_string *old_ns;
 821
 822                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 823                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 824                                         lockdep_is_held(&ci->i_ceph_lock));
 825                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 826
 827                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 828                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 829
 830                 pool_ns = old_ns;
 831
 832                 queue_trunc = ceph_fill_file_size(inode, issued,
 833                                         le32_to_cpu(info->truncate_seq),
 834                                         le64_to_cpu(info->truncate_size),
 835                                         le64_to_cpu(info->size));
 836                 /* only update max_size on auth cap */
 837                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 838                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 839                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 840                                         le64_to_cpu(info->max_size));
 841                         ci->i_max_size = le64_to_cpu(info->max_size);
 842                 }
 843         }
 844
 845         /* xattrs */
 846         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 847         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 848             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 849                 if (ci->i_xattrs.blob)
 850                         ceph_buffer_put(ci->i_xattrs.blob);
 851                 ci->i_xattrs.blob = xattr_blob;
 852                 if (xattr_blob)
 853                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 854                                iinfo->xattr_data, iinfo->xattr_len);
 855                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 856                 ceph_forget_all_cached_acls(inode);
 857                 xattr_blob = NULL;
 858         }
 859
 860         inode->i_mapping->a_ops = &ceph_aops;
 861
 862         switch (inode->i_mode & S_IFMT) {
 863         case S_IFIFO:
 864         case S_IFBLK:
 865         case S_IFCHR:
 866         case S_IFSOCK:
 867                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 868                 inode->i_op = &ceph_file_iops;
 869                 break;
 870         case S_IFREG:
 871                 inode->i_op = &ceph_file_iops;
 872                 inode->i_fop = &ceph_file_fops;
 873                 break;
 874         case S_IFLNK:
 875                 inode->i_op = &ceph_symlink_iops;
 876                 if (!ci->i_symlink) {
 877                         u32 symlen = iinfo->symlink_len;
 878                         char *sym;
 879
 880                         spin_unlock(&ci->i_ceph_lock);
 881
 882                         if (symlen != i_size_read(inode)) {
 883                                 pr_err("fill_inode %llx.%llx BAD symlink "
 884                                         "size %lld\n", ceph_vinop(inode),
 885                                         i_size_read(inode));
 886                                 i_size_write(inode, symlen);
 887                                 inode->i_blocks = calc_inode_blocks(symlen);
 888                         }
 889
 890                         err = -ENOMEM;
 891                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 892                         if (!sym)
 893                                 goto out;
 894
 895                         spin_lock(&ci->i_ceph_lock);
 896                         if (!ci->i_symlink)
 897                                 ci->i_symlink = sym;
 898                         else
 899                                 kfree(sym); /* lost a race */
 900                 }
 901                 inode->i_link = ci->i_symlink;
 902                 break;
 903         case S_IFDIR:
 904                 inode->i_op = &ceph_dir_iops;
 905                 inode->i_fop = &ceph_dir_fops;
 906
 907                 ci->i_dir_layout = iinfo->dir_layout;
 908
 909                 ci->i_files = le64_to_cpu(info->files);
 910                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 911                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 912                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 913                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 914                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 915                 break;
 916         default:
 917                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 918                        ceph_vinop(inode), inode->i_mode);
 919         }
 920
 921         /* were we issued a capability? */
 922         if (info->cap.caps) {
 923                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 924                         unsigned caps = le32_to_cpu(info->cap.caps);
 925                         ceph_add_cap(inode, session,
 926                                      le64_to_cpu(info->cap.cap_id),
 927                                      cap_fmode, caps,
 928                                      le32_to_cpu(info->cap.wanted),
 929                                      le32_to_cpu(info->cap.seq),
 930                                      le32_to_cpu(info->cap.mseq),
 931                                      le64_to_cpu(info->cap.realm),
 932                                      info->cap.flags, &new_cap);
 933
 934                         /* set dir completion flag? */
 935                         if (S_ISDIR(inode->i_mode) &&
 936                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 937                             (caps & CEPH_CAP_FILE_SHARED) &&
 938                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 939                             !__ceph_dir_is_complete(ci)) {
 940                                 dout(" marking %p complete (empty)\n", inode);
 941                                 i_size_write(inode, 0);
 942                                 __ceph_dir_set_complete(ci,
 943                                         atomic64_read(&ci->i_release_count),
 944                                         atomic64_read(&ci->i_ordered_count));
 945                         }
 946
 947                         wake = true;
 948                 } else {
 949                         dout(" %p got snap_caps %s\n", inode,
 950                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 951                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 952                         if (cap_fmode >= 0)
 953                                 __ceph_get_fmode(ci, cap_fmode);
 954                 }
 955         } else if (cap_fmode >= 0) {
 956                 pr_warn("mds issued no caps on %llx.%llx\n",
 957                            ceph_vinop(inode));
 958                 __ceph_get_fmode(ci, cap_fmode);
 959         }
 960
 961         if (iinfo->inline_version > 0 &&
 962             iinfo->inline_version >= ci->i_inline_version) {
 963                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 964                 ci->i_inline_version = iinfo->inline_version;
 965                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 966                     (locked_page ||
 967                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 968                         fill_inline = true;
 969         }
 970
 971         spin_unlock(&ci->i_ceph_lock);
 972
 973         if (fill_inline)
 974                 ceph_fill_inline_data(inode, locked_page,
 975                                       iinfo->inline_data, iinfo->inline_len);
 976
 977         if (wake)
 978                 wake_up_all(&ci->i_cap_wq);
 979
 980         /* queue truncate if we saw i_size decrease */
 981         if (queue_trunc)
 982                 ceph_queue_vmtruncate(inode);
 983
 984         /* populate frag tree */
 985         if (S_ISDIR(inode->i_mode))
 986                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 987
 988         /* update delegation info? */
 989         if (dirinfo)
 990                 ceph_fill_dirfrag(inode, dirinfo);
 991
 992         err = 0;
 993 out:
 994         if (new_cap)
 995                 ceph_put_cap(mdsc, new_cap);
 996         if (xattr_blob)
 997                 ceph_buffer_put(xattr_blob);
 998         ceph_put_string(pool_ns);
 999         return err;
1000 }
1001
1002 /*
1003  * caller should hold session s_mutex.
1004  */
1005 static void update_dentry_lease(struct dentry *dentry,
1006                                 struct ceph_mds_reply_lease *lease,
1007                                 struct ceph_mds_session *session,
1008                                 unsigned long from_time,
1009                                 struct ceph_vino *tgt_vino,
1010                                 struct ceph_vino *dir_vino)
1011 {
1012         struct ceph_dentry_info *di = ceph_dentry(dentry);
1013         long unsigned duration = le32_to_cpu(lease->duration_ms);
1014         long unsigned ttl = from_time + (duration * HZ) / 1000;
1015         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1016         struct inode *dir;
1017         struct ceph_mds_session *old_lease_session = NULL;
1018
1019         /*
1020          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1021          * we expect a negative dentry.
1022          */
1023         if (!tgt_vino && d_really_is_positive(dentry))
1024                 return;
1025
1026         if (tgt_vino && (d_really_is_negative(dentry) ||
1027                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1028                 return;
1029
1030         spin_lock(&dentry->d_lock);
1031         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1032              dentry, duration, ttl);
1033
1034         dir = d_inode(dentry->d_parent);
1035
1036         /* make sure parent matches dir_vino */
1037         if (!ceph_ino_compare(dir, dir_vino))
1038                 goto out_unlock;
1039
1040         /* only track leases on regular dentries */
1041         if (ceph_snap(dir) != CEPH_NOSNAP)
1042                 goto out_unlock;
1043
1044         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1045
1046         if (duration == 0)
1047                 goto out_unlock;
1048
1049         if (di->lease_gen == session->s_cap_gen &&
1050             time_before(ttl, di->time))
1051                 goto out_unlock;  /* we already have a newer lease. */
1052
1053         if (di->lease_session && di->lease_session != session) {
1054                 old_lease_session = di->lease_session;
1055                 di->lease_session = NULL;
1056         }
1057
1058         ceph_dentry_lru_touch(dentry);
1059
1060         if (!di->lease_session)
1061                 di->lease_session = ceph_get_mds_session(session);
1062         di->lease_gen = session->s_cap_gen;
1063         di->lease_seq = le32_to_cpu(lease->seq);
1064         di->lease_renew_after = half_ttl;
1065         di->lease_renew_from = 0;
1066         di->time = ttl;
1067 out_unlock:
1068         spin_unlock(&dentry->d_lock);
1069         if (old_lease_session)
1070                 ceph_put_mds_session(old_lease_session);
1071 }
1072
1073 /*
1074  * splice a dentry to an inode.
1075  * caller must hold directory i_mutex for this to be safe.
1076  */
1077 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1078 {
1079         struct dentry *realdn;
1080
1081         BUG_ON(d_inode(dn));
1082
1083         /* dn must be unhashed */
1084         if (!d_unhashed(dn))
1085                 d_drop(dn);
1086         realdn = d_splice_alias(in, dn);
1087         if (IS_ERR(realdn)) {
1088                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1089                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1090                 dn = realdn; /* note realdn contains the error */
1091                 goto out;
1092         } else if (realdn) {
1093                 dout("dn %p (%d) spliced with %p (%d) "
1094                      "inode %p ino %llx.%llx\n",
1095                      dn, d_count(dn),
1096                      realdn, d_count(realdn),
1097                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1098                 dput(dn);
1099                 dn = realdn;
1100         } else {
1101                 BUG_ON(!ceph_dentry(dn));
1102                 dout("dn %p attached to %p ino %llx.%llx\n",
1103                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1104         }
1105 out:
1106         return dn;
1107 }
1108
1109 /*
1110  * Incorporate results into the local cache.  This is either just
1111  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1112  * after a lookup).
1113  *
1114  * A reply may contain
1115  *         a directory inode along with a dentry.
1116  *  and/or a target inode
1117  *
1118  * Called with snap_rwsem (read).
1119  */
1120 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1121 {
1122         struct ceph_mds_session *session = req->r_session;
1123         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1124         struct inode *in = NULL;
1125         struct ceph_vino tvino, dvino;
1126         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1127         int err = 0;
1128
1129         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1130              rinfo->head->is_dentry, rinfo->head->is_target);
1131
1132         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1133                 dout("fill_trace reply is empty!\n");
1134                 if (rinfo->head->result == 0 && req->r_parent)
1135                         ceph_invalidate_dir_request(req);
1136                 return 0;
1137         }
1138
1139         if (rinfo->head->is_dentry) {
1140                 struct inode *dir = req->r_parent;
1141
1142                 if (dir) {
1143                         err = fill_inode(dir, NULL,
1144                                          &rinfo->diri, rinfo->dirfrag,
1145                                          session, req->r_request_started, -1,
1146                                          &req->r_caps_reservation);
1147                         if (err < 0)
1148                                 goto done;
1149                 } else {
1150                         WARN_ON_ONCE(1);
1151                 }
1152
1153                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1154                         struct qstr dname;
1155                         struct dentry *dn, *parent;
1156
1157                         BUG_ON(!rinfo->head->is_target);
1158                         BUG_ON(req->r_dentry);
1159
1160                         parent = d_find_any_alias(dir);
1161                         BUG_ON(!parent);
1162
1163                         dname.name = rinfo->dname;
1164                         dname.len = rinfo->dname_len;
1165                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1166                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1167                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1168 retry_lookup:
1169                         dn = d_lookup(parent, &dname);
1170                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1171                              parent, dname.len, dname.name, dn);
1172
1173                         if (!dn) {
1174                                 dn = d_alloc(parent, &dname);
1175                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1176                                      dname.len, dname.name, dn);
1177                                 if (!dn) {
1178                                         dput(parent);
1179                                         err = -ENOMEM;
1180                                         goto done;
1181                                 }
1182                                 err = 0;
1183                         } else if (d_really_is_positive(dn) &&
1184                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1185                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1186                                 dout(" dn %p points to wrong inode %p\n",
1187                                      dn, d_inode(dn));
1188                                 d_delete(dn);
1189                                 dput(dn);
1190                                 goto retry_lookup;
1191                         }
1192
1193                         req->r_dentry = dn;
1194                         dput(parent);
1195                 }
1196         }
1197
1198         if (rinfo->head->is_target) {
1199                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1200                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1201
1202                 in = ceph_get_inode(sb, tvino);
1203                 if (IS_ERR(in)) {
1204                         err = PTR_ERR(in);
1205                         goto done;
1206                 }
1207                 req->r_target_inode = in;
1208
1209                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1210                                 session, req->r_request_started,
1211                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1212                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1213                                 &req->r_caps_reservation);
1214                 if (err < 0) {
1215                         pr_err("fill_inode badness %p %llx.%llx\n",
1216                                 in, ceph_vinop(in));
1217                         goto done;
1218                 }
1219         }
1220
1221         /*
1222          * ignore null lease/binding on snapdir ENOENT, or else we
1223          * will have trouble splicing in the virtual snapdir later
1224          */
1225         if (rinfo->head->is_dentry &&
1226             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1227             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1228             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1229                                                fsc->mount_options->snapdir_name,
1230                                                req->r_dentry->d_name.len))) {
1231                 /*
1232                  * lookup link rename   : null -> possibly existing inode
1233                  * mknod symlink mkdir  : null -> new inode
1234                  * unlink               : linked -> null
1235                  */
1236                 struct inode *dir = req->r_parent;
1237                 struct dentry *dn = req->r_dentry;
1238                 bool have_dir_cap, have_lease;
1239
1240                 BUG_ON(!dn);
1241                 BUG_ON(!dir);
1242                 BUG_ON(d_inode(dn->d_parent) != dir);
1243
1244                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1245                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1246
1247                 BUG_ON(ceph_ino(dir) != dvino.ino);
1248                 BUG_ON(ceph_snap(dir) != dvino.snap);
1249
1250                 /* do we have a lease on the whole dir? */
1251                 have_dir_cap =
1252                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1253                          CEPH_CAP_FILE_SHARED);
1254
1255                 /* do we have a dn lease? */
1256                 have_lease = have_dir_cap ||
1257                         le32_to_cpu(rinfo->dlease->duration_ms);
1258                 if (!have_lease)
1259                         dout("fill_trace  no dentry lease or dir cap\n");
1260
1261                 /* rename? */
1262                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1263                         struct inode *olddir = req->r_old_dentry_dir;
1264                         BUG_ON(!olddir);
1265
1266                         dout(" src %p '%pd' dst %p '%pd'\n",
1267                              req->r_old_dentry,
1268                              req->r_old_dentry,
1269                              dn, dn);
1270                         dout("fill_trace doing d_move %p -> %p\n",
1271                              req->r_old_dentry, dn);
1272
1273                         /* d_move screws up sibling dentries' offsets */
1274                         ceph_dir_clear_ordered(dir);
1275                         ceph_dir_clear_ordered(olddir);
1276
1277                         d_move(req->r_old_dentry, dn);
1278                         dout(" src %p '%pd' dst %p '%pd'\n",
1279                              req->r_old_dentry,
1280                              req->r_old_dentry,
1281                              dn, dn);
1282
1283                         /* ensure target dentry is invalidated, despite
1284                            rehashing bug in vfs_rename_dir */
1285                         ceph_invalidate_dentry_lease(dn);
1286
1287                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1288                              ceph_dentry(req->r_old_dentry)->offset);
1289
1290                         dn = req->r_old_dentry;  /* use old_dentry */
1291                 }
1292
1293                 /* null dentry? */
1294                 if (!rinfo->head->is_target) {
1295                         dout("fill_trace null dentry\n");
1296                         if (d_really_is_positive(dn)) {
1297                                 ceph_dir_clear_ordered(dir);
1298                                 dout("d_delete %p\n", dn);
1299                                 d_delete(dn);
1300                         } else if (have_lease) {
1301                                 if (d_unhashed(dn))
1302                                         d_add(dn, NULL);
1303                                 update_dentry_lease(dn, rinfo->dlease,
1304                                                     session,
1305                                                     req->r_request_started,
1306                                                     NULL, &dvino);
1307                         }
1308                         goto done;
1309                 }
1310
1311                 /* attach proper inode */
1312                 if (d_really_is_negative(dn)) {
1313                         ceph_dir_clear_ordered(dir);
1314                         ihold(in);
1315                         dn = splice_dentry(dn, in);
1316                         if (IS_ERR(dn)) {
1317                                 err = PTR_ERR(dn);
1318                                 goto done;
1319                         }
1320                         req->r_dentry = dn;  /* may have spliced */
1321                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1322                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1323                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1324                              ceph_vinop(in));
1325                         d_invalidate(dn);
1326                         have_lease = false;
1327                 }
1328
1329                 if (have_lease) {
1330                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1331                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1332                         update_dentry_lease(dn, rinfo->dlease, session,
1333                                             req->r_request_started,
1334                                             &tvino, &dvino);
1335                 }
1336                 dout(" final dn %p\n", dn);
1337         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1338                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1339                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1340                 struct dentry *dn = req->r_dentry;
1341                 struct inode *dir = req->r_parent;
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         } else if (rinfo->head->is_dentry) {
1357                 struct ceph_vino *ptvino = NULL;
1358
1359                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1360                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1361                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1362                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1363
1364                         if (rinfo->head->is_target) {
1365                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1366                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1367                                 ptvino = &tvino;
1368                         }
1369
1370                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1371                                 session, req->r_request_started, ptvino,
1372                                 &dvino);
1373                 } else {
1374                         dout("%s: no dentry lease or dir cap\n", __func__);
1375                 }
1376         }
1377 done:
1378         dout("fill_trace done err=%d\n", err);
1379         return err;
1380 }
1381
1382 /*
1383  * Prepopulate our cache with readdir results, leases, etc.
1384  */
1385 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1386                                            struct ceph_mds_session *session)
1387 {
1388         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1389         int i, err = 0;
1390
1391         for (i = 0; i < rinfo->dir_nr; i++) {
1392                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1393                 struct ceph_vino vino;
1394                 struct inode *in;
1395                 int rc;
1396
1397                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1398                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1399
1400                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1401                 if (IS_ERR(in)) {
1402                         err = PTR_ERR(in);
1403                         dout("new_inode badness got %d\n", err);
1404                         continue;
1405                 }
1406                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1407                                 req->r_request_started, -1,
1408                                 &req->r_caps_reservation);
1409                 if (rc < 0) {
1410                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1411                         err = rc;
1412                 }
1413                 iput(in);
1414         }
1415
1416         return err;
1417 }
1418
1419 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1420 {
1421         if (ctl->page) {
1422                 kunmap(ctl->page);
1423                 put_page(ctl->page);
1424                 ctl->page = NULL;
1425         }
1426 }
1427
1428 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1429                               struct ceph_readdir_cache_control *ctl,
1430                               struct ceph_mds_request *req)
1431 {
1432         struct ceph_inode_info *ci = ceph_inode(dir);
1433         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1434         unsigned idx = ctl->index % nsize;
1435         pgoff_t pgoff = ctl->index / nsize;
1436
1437         if (!ctl->page || pgoff != page_index(ctl->page)) {
1438                 ceph_readdir_cache_release(ctl);
1439                 if (idx == 0)
1440                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1441                 else
1442                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1443                 if (!ctl->page) {
1444                         ctl->index = -1;
1445                         return idx == 0 ? -ENOMEM : 0;
1446                 }
1447                 /* reading/filling the cache are serialized by
1448                  * i_mutex, no need to use page lock */
1449                 unlock_page(ctl->page);
1450                 ctl->dentries = kmap(ctl->page);
1451                 if (idx == 0)
1452                         memset(ctl->dentries, 0, PAGE_SIZE);
1453         }
1454
1455         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1456             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1457                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1458                 ctl->dentries[idx] = dn;
1459                 ctl->index++;
1460         } else {
1461                 dout("disable readdir cache\n");
1462                 ctl->index = -1;
1463         }
1464         return 0;
1465 }
1466
1467 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1468                              struct ceph_mds_session *session)
1469 {
1470         struct dentry *parent = req->r_dentry;
1471         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1472         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1473         struct qstr dname;
1474         struct dentry *dn;
1475         struct inode *in;
1476         int err = 0, skipped = 0, ret, i;
1477         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1478         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1479         u32 last_hash = 0;
1480         u32 fpos_offset;
1481         struct ceph_readdir_cache_control cache_ctl = {};
1482
1483         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1484                 return readdir_prepopulate_inodes_only(req, session);
1485
1486         if (rinfo->hash_order) {
1487                 if (req->r_path2) {
1488                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1489                                                   req->r_path2,
1490                                                   strlen(req->r_path2));
1491                         last_hash = ceph_frag_value(last_hash);
1492                 } else if (rinfo->offset_hash) {
1493                         /* mds understands offset_hash */
1494                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1495                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1496                 }
1497         }
1498
1499         if (rinfo->dir_dir &&
1500             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1501                 dout("readdir_prepopulate got new frag %x -> %x\n",
1502                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1503                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1504                 if (!rinfo->hash_order)
1505                         req->r_readdir_offset = 2;
1506         }
1507
1508         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1509                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1510                      rinfo->dir_nr, parent);
1511         } else {
1512                 dout("readdir_prepopulate %d items under dn %p\n",
1513                      rinfo->dir_nr, parent);
1514                 if (rinfo->dir_dir)
1515                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1516
1517                 if (ceph_frag_is_leftmost(frag) &&
1518                     req->r_readdir_offset == 2 &&
1519                     !(rinfo->hash_order && last_hash)) {
1520                         /* note dir version at start of readdir so we can
1521                          * tell if any dentries get dropped */
1522                         req->r_dir_release_cnt =
1523                                 atomic64_read(&ci->i_release_count);
1524                         req->r_dir_ordered_cnt =
1525                                 atomic64_read(&ci->i_ordered_count);
1526                         req->r_readdir_cache_idx = 0;
1527                 }
1528         }
1529
1530         cache_ctl.index = req->r_readdir_cache_idx;
1531         fpos_offset = req->r_readdir_offset;
1532
1533         /* FIXME: release caps/leases if error occurs */
1534         for (i = 0; i < rinfo->dir_nr; i++) {
1535                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1536                 struct ceph_vino tvino, dvino;
1537
1538                 dname.name = rde->name;
1539                 dname.len = rde->name_len;
1540                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1541
1542                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1543                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1544
1545                 if (rinfo->hash_order) {
1546                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1547                                                  rde->name, rde->name_len);
1548                         hash = ceph_frag_value(hash);
1549                         if (hash != last_hash)
1550                                 fpos_offset = 2;
1551                         last_hash = hash;
1552                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1553                 } else {
1554                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1555                 }
1556
1557 retry_lookup:
1558                 dn = d_lookup(parent, &dname);
1559                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1560                      parent, dname.len, dname.name, dn);
1561
1562                 if (!dn) {
1563                         dn = d_alloc(parent, &dname);
1564                         dout("d_alloc %p '%.*s' = %p\n", parent,
1565                              dname.len, dname.name, dn);
1566                         if (!dn) {
1567                                 dout("d_alloc badness\n");
1568                                 err = -ENOMEM;
1569                                 goto out;
1570                         }
1571                 } else if (d_really_is_positive(dn) &&
1572                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1573                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1574                         dout(" dn %p points to wrong inode %p\n",
1575                              dn, d_inode(dn));
1576                         d_delete(dn);
1577                         dput(dn);
1578                         goto retry_lookup;
1579                 }
1580
1581                 /* inode */
1582                 if (d_really_is_positive(dn)) {
1583                         in = d_inode(dn);
1584                 } else {
1585                         in = ceph_get_inode(parent->d_sb, tvino);
1586                         if (IS_ERR(in)) {
1587                                 dout("new_inode badness\n");
1588                                 d_drop(dn);
1589                                 dput(dn);
1590                                 err = PTR_ERR(in);
1591                                 goto out;
1592                         }
1593                 }
1594
1595                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1596                                  req->r_request_started, -1,
1597                                  &req->r_caps_reservation);
1598                 if (ret < 0) {
1599                         pr_err("fill_inode badness on %p\n", in);
1600                         if (d_really_is_negative(dn))
1601                                 iput(in);
1602                         d_drop(dn);
1603                         err = ret;
1604                         goto next_item;
1605                 }
1606
1607                 if (d_really_is_negative(dn)) {
1608                         struct dentry *realdn;
1609
1610                         if (ceph_security_xattr_deadlock(in)) {
1611                                 dout(" skip splicing dn %p to inode %p"
1612                                      " (security xattr deadlock)\n", dn, in);
1613                                 iput(in);
1614                                 skipped++;
1615                                 goto next_item;
1616                         }
1617
1618                         realdn = splice_dentry(dn, in);
1619                         if (IS_ERR(realdn)) {
1620                                 err = PTR_ERR(realdn);
1621                                 d_drop(dn);
1622                                 dn = NULL;
1623                                 goto next_item;
1624                         }
1625                         dn = realdn;
1626                 }
1627
1628                 ceph_dentry(dn)->offset = rde->offset;
1629
1630                 dvino = ceph_vino(d_inode(parent));
1631                 update_dentry_lease(dn, rde->lease, req->r_session,
1632                                     req->r_request_started, &tvino, &dvino);
1633
1634                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1635                         ret = fill_readdir_cache(d_inode(parent), dn,
1636                                                  &cache_ctl, req);
1637                         if (ret < 0)
1638                                 err = ret;
1639                 }
1640 next_item:
1641                 if (dn)
1642                         dput(dn);
1643         }
1644 out:
1645         if (err == 0 && skipped == 0) {
1646                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1647                 req->r_readdir_cache_idx = cache_ctl.index;
1648         }
1649         ceph_readdir_cache_release(&cache_ctl);
1650         dout("readdir_prepopulate done\n");
1651         return err;
1652 }
1653
1654 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1655 {
1656         struct ceph_inode_info *ci = ceph_inode(inode);
1657         bool ret;
1658
1659         spin_lock(&ci->i_ceph_lock);
1660         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1661         i_size_write(inode, size);
1662         inode->i_blocks = calc_inode_blocks(size);
1663
1664         ret = __ceph_should_report_size(ci);
1665
1666         spin_unlock(&ci->i_ceph_lock);
1667         return ret;
1668 }
1669
1670 /*
1671  * Write back inode data in a worker thread.  (This can't be done
1672  * in the message handler context.)
1673  */
1674 void ceph_queue_writeback(struct inode *inode)
1675 {
1676         ihold(inode);
1677         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1678                        &ceph_inode(inode)->i_wb_work)) {
1679                 dout("ceph_queue_writeback %p\n", inode);
1680         } else {
1681                 dout("ceph_queue_writeback %p failed\n", inode);
1682                 iput(inode);
1683         }
1684 }
1685
1686 static void ceph_writeback_work(struct work_struct *work)
1687 {
1688         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1689                                                   i_wb_work);
1690         struct inode *inode = &ci->vfs_inode;
1691
1692         dout("writeback %p\n", inode);
1693         filemap_fdatawrite(&inode->i_data);
1694         iput(inode);
1695 }
1696
1697 /*
1698  * queue an async invalidation
1699  */
1700 void ceph_queue_invalidate(struct inode *inode)
1701 {
1702         ihold(inode);
1703         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1704                        &ceph_inode(inode)->i_pg_inv_work)) {
1705                 dout("ceph_queue_invalidate %p\n", inode);
1706         } else {
1707                 dout("ceph_queue_invalidate %p failed\n", inode);
1708                 iput(inode);
1709         }
1710 }
1711
1712 /*
1713  * Invalidate inode pages in a worker thread.  (This can't be done
1714  * in the message handler context.)
1715  */
1716 static void ceph_invalidate_work(struct work_struct *work)
1717 {
1718         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1719                                                   i_pg_inv_work);
1720         struct inode *inode = &ci->vfs_inode;
1721         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1722         u32 orig_gen;
1723         int check = 0;
1724
1725         mutex_lock(&ci->i_truncate_mutex);
1726
1727         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1728                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1729                                     inode, ceph_ino(inode));
1730                 mapping_set_error(inode->i_mapping, -EIO);
1731                 truncate_pagecache(inode, 0);
1732                 mutex_unlock(&ci->i_truncate_mutex);
1733                 goto out;
1734         }
1735
1736         spin_lock(&ci->i_ceph_lock);
1737         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1738              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1739         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1740                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1741                         check = 1;
1742                 spin_unlock(&ci->i_ceph_lock);
1743                 mutex_unlock(&ci->i_truncate_mutex);
1744                 goto out;
1745         }
1746         orig_gen = ci->i_rdcache_gen;
1747         spin_unlock(&ci->i_ceph_lock);
1748
1749         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1750                 pr_err("invalidate_pages %p fails\n", inode);
1751         }
1752
1753         spin_lock(&ci->i_ceph_lock);
1754         if (orig_gen == ci->i_rdcache_gen &&
1755             orig_gen == ci->i_rdcache_revoking) {
1756                 dout("invalidate_pages %p gen %d successful\n", inode,
1757                      ci->i_rdcache_gen);
1758                 ci->i_rdcache_revoking--;
1759                 check = 1;
1760         } else {
1761                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1762                      inode, orig_gen, ci->i_rdcache_gen,
1763                      ci->i_rdcache_revoking);
1764                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1765                         check = 1;
1766         }
1767         spin_unlock(&ci->i_ceph_lock);
1768         mutex_unlock(&ci->i_truncate_mutex);
1769 out:
1770         if (check)
1771                 ceph_check_caps(ci, 0, NULL);
1772         iput(inode);
1773 }
1774
1775
1776 /*
1777  * called by trunc_wq;
1778  *
1779  * We also truncate in a separate thread as well.
1780  */
1781 static void ceph_vmtruncate_work(struct work_struct *work)
1782 {
1783         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1784                                                   i_vmtruncate_work);
1785         struct inode *inode = &ci->vfs_inode;
1786
1787         dout("vmtruncate_work %p\n", inode);
1788         __ceph_do_pending_vmtruncate(inode);
1789         iput(inode);
1790 }
1791
1792 /*
1793  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1794  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1795  */
1796 void ceph_queue_vmtruncate(struct inode *inode)
1797 {
1798         struct ceph_inode_info *ci = ceph_inode(inode);
1799
1800         ihold(inode);
1801
1802         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1803                        &ci->i_vmtruncate_work)) {
1804                 dout("ceph_queue_vmtruncate %p\n", inode);
1805         } else {
1806                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1807                      inode, ci->i_truncate_pending);
1808                 iput(inode);
1809         }
1810 }
1811
1812 /*
1813  * Make sure any pending truncation is applied before doing anything
1814  * that may depend on it.
1815  */
1816 void __ceph_do_pending_vmtruncate(struct inode *inode)
1817 {
1818         struct ceph_inode_info *ci = ceph_inode(inode);
1819         u64 to;
1820         int wrbuffer_refs, finish = 0;
1821
1822         mutex_lock(&ci->i_truncate_mutex);
1823 retry:
1824         spin_lock(&ci->i_ceph_lock);
1825         if (ci->i_truncate_pending == 0) {
1826                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1827                 spin_unlock(&ci->i_ceph_lock);
1828                 mutex_unlock(&ci->i_truncate_mutex);
1829                 return;
1830         }
1831
1832         /*
1833          * make sure any dirty snapped pages are flushed before we
1834          * possibly truncate them.. so write AND block!
1835          */
1836         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1837                 struct ceph_cap_snap *capsnap;
1838                 to = ci->i_truncate_size;
1839                 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1840                         // MDS should have revoked Frw caps
1841                         WARN_ON_ONCE(capsnap->writing);
1842                         if (capsnap->dirty_pages && capsnap->size > to)
1843                                 to = capsnap->size;
1844                 }
1845                 spin_unlock(&ci->i_ceph_lock);
1846                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1847                      inode);
1848
1849                 truncate_pagecache(inode, to);
1850
1851                 filemap_write_and_wait_range(&inode->i_data, 0,
1852                                              inode->i_sb->s_maxbytes);
1853                 goto retry;
1854         }
1855
1856         /* there should be no reader or writer */
1857         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1858
1859         to = ci->i_truncate_size;
1860         wrbuffer_refs = ci->i_wrbuffer_ref;
1861         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1862              ci->i_truncate_pending, to);
1863         spin_unlock(&ci->i_ceph_lock);
1864
1865         truncate_pagecache(inode, to);
1866
1867         spin_lock(&ci->i_ceph_lock);
1868         if (to == ci->i_truncate_size) {
1869                 ci->i_truncate_pending = 0;
1870                 finish = 1;
1871         }
1872         spin_unlock(&ci->i_ceph_lock);
1873         if (!finish)
1874                 goto retry;
1875
1876         mutex_unlock(&ci->i_truncate_mutex);
1877
1878         if (wrbuffer_refs == 0)
1879                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1880
1881         wake_up_all(&ci->i_cap_wq);
1882 }
1883
1884 /*
1885  * symlinks
1886  */
1887 static const struct inode_operations ceph_symlink_iops = {
1888         .get_link = simple_get_link,
1889         .setattr = ceph_setattr,
1890         .getattr = ceph_getattr,
1891         .listxattr = ceph_listxattr,
1892 };
1893
1894 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1895 {
1896         struct ceph_inode_info *ci = ceph_inode(inode);
1897         const unsigned int ia_valid = attr->ia_valid;
1898         struct ceph_mds_request *req;
1899         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1900         struct ceph_cap_flush *prealloc_cf;
1901         int issued;
1902         int release = 0, dirtied = 0;
1903         int mask = 0;
1904         int err = 0;
1905         int inode_dirty_flags = 0;
1906         bool lock_snap_rwsem = false;
1907
1908         prealloc_cf = ceph_alloc_cap_flush();
1909         if (!prealloc_cf)
1910                 return -ENOMEM;
1911
1912         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1913                                        USE_AUTH_MDS);
1914         if (IS_ERR(req)) {
1915                 ceph_free_cap_flush(prealloc_cf);
1916                 return PTR_ERR(req);
1917         }
1918
1919         spin_lock(&ci->i_ceph_lock);
1920         issued = __ceph_caps_issued(ci, NULL);
1921
1922         if (!ci->i_head_snapc &&
1923             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1924                 lock_snap_rwsem = true;
1925                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1926                         spin_unlock(&ci->i_ceph_lock);
1927                         down_read(&mdsc->snap_rwsem);
1928                         spin_lock(&ci->i_ceph_lock);
1929                         issued = __ceph_caps_issued(ci, NULL);
1930                 }
1931         }
1932
1933         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1934
1935         if (ia_valid & ATTR_UID) {
1936                 dout("setattr %p uid %d -> %d\n", inode,
1937                      from_kuid(&init_user_ns, inode->i_uid),
1938                      from_kuid(&init_user_ns, attr->ia_uid));
1939                 if (issued & CEPH_CAP_AUTH_EXCL) {
1940                         inode->i_uid = attr->ia_uid;
1941                         dirtied |= CEPH_CAP_AUTH_EXCL;
1942                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1943                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1944                         req->r_args.setattr.uid = cpu_to_le32(
1945                                 from_kuid(&init_user_ns, attr->ia_uid));
1946                         mask |= CEPH_SETATTR_UID;
1947                         release |= CEPH_CAP_AUTH_SHARED;
1948                 }
1949         }
1950         if (ia_valid & ATTR_GID) {
1951                 dout("setattr %p gid %d -> %d\n", inode,
1952                      from_kgid(&init_user_ns, inode->i_gid),
1953                      from_kgid(&init_user_ns, attr->ia_gid));
1954                 if (issued & CEPH_CAP_AUTH_EXCL) {
1955                         inode->i_gid = attr->ia_gid;
1956                         dirtied |= CEPH_CAP_AUTH_EXCL;
1957                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1958                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1959                         req->r_args.setattr.gid = cpu_to_le32(
1960                                 from_kgid(&init_user_ns, attr->ia_gid));
1961                         mask |= CEPH_SETATTR_GID;
1962                         release |= CEPH_CAP_AUTH_SHARED;
1963                 }
1964         }
1965         if (ia_valid & ATTR_MODE) {
1966                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1967                      attr->ia_mode);
1968                 if (issued & CEPH_CAP_AUTH_EXCL) {
1969                         inode->i_mode = attr->ia_mode;
1970                         dirtied |= CEPH_CAP_AUTH_EXCL;
1971                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1972                            attr->ia_mode != inode->i_mode) {
1973                         inode->i_mode = attr->ia_mode;
1974                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1975                         mask |= CEPH_SETATTR_MODE;
1976                         release |= CEPH_CAP_AUTH_SHARED;
1977                 }
1978         }
1979
1980         if (ia_valid & ATTR_ATIME) {
1981                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1982                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1983                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1984                 if (issued & CEPH_CAP_FILE_EXCL) {
1985                         ci->i_time_warp_seq++;
1986                         inode->i_atime = attr->ia_atime;
1987                         dirtied |= CEPH_CAP_FILE_EXCL;
1988                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1989                            timespec_compare(&inode->i_atime,
1990                                             &attr->ia_atime) < 0) {
1991                         inode->i_atime = attr->ia_atime;
1992                         dirtied |= CEPH_CAP_FILE_WR;
1993                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1994                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1995                         ceph_encode_timespec(&req->r_args.setattr.atime,
1996                                              &attr->ia_atime);
1997                         mask |= CEPH_SETATTR_ATIME;
1998                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1999                                 CEPH_CAP_FILE_WR;
2000                 }
2001         }
2002         if (ia_valid & ATTR_MTIME) {
2003                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
2004                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2005                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2006                 if (issued & CEPH_CAP_FILE_EXCL) {
2007                         ci->i_time_warp_seq++;
2008                         inode->i_mtime = attr->ia_mtime;
2009                         dirtied |= CEPH_CAP_FILE_EXCL;
2010                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2011                            timespec_compare(&inode->i_mtime,
2012                                             &attr->ia_mtime) < 0) {
2013                         inode->i_mtime = attr->ia_mtime;
2014                         dirtied |= CEPH_CAP_FILE_WR;
2015                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2016                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2017                         ceph_encode_timespec(&req->r_args.setattr.mtime,
2018                                              &attr->ia_mtime);
2019                         mask |= CEPH_SETATTR_MTIME;
2020                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2021                                 CEPH_CAP_FILE_WR;
2022                 }
2023         }
2024         if (ia_valid & ATTR_SIZE) {
2025                 dout("setattr %p size %lld -> %lld\n", inode,
2026                      inode->i_size, attr->ia_size);
2027                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2028                     attr->ia_size > inode->i_size) {
2029                         i_size_write(inode, attr->ia_size);
2030                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2031                         ci->i_reported_size = attr->ia_size;
2032                         dirtied |= CEPH_CAP_FILE_EXCL;
2033                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2034                            attr->ia_size != inode->i_size) {
2035                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2036                         req->r_args.setattr.old_size =
2037                                 cpu_to_le64(inode->i_size);
2038                         mask |= CEPH_SETATTR_SIZE;
2039                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2040                                 CEPH_CAP_FILE_WR;
2041                 }
2042         }
2043
2044         /* these do nothing */
2045         if (ia_valid & ATTR_CTIME) {
2046                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2047                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2048                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2049                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2050                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2051                      only ? "ctime only" : "ignored");
2052                 if (only) {
2053                         /*
2054                          * if kernel wants to dirty ctime but nothing else,
2055                          * we need to choose a cap to dirty under, or do
2056                          * a almost-no-op setattr
2057                          */
2058                         if (issued & CEPH_CAP_AUTH_EXCL)
2059                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2060                         else if (issued & CEPH_CAP_FILE_EXCL)
2061                                 dirtied |= CEPH_CAP_FILE_EXCL;
2062                         else if (issued & CEPH_CAP_XATTR_EXCL)
2063                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2064                         else
2065                                 mask |= CEPH_SETATTR_CTIME;
2066                 }
2067         }
2068         if (ia_valid & ATTR_FILE)
2069                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2070
2071         if (dirtied) {
2072                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2073                                                            &prealloc_cf);
2074                 inode->i_ctime = attr->ia_ctime;
2075         }
2076
2077         release &= issued;
2078         spin_unlock(&ci->i_ceph_lock);
2079         if (lock_snap_rwsem)
2080                 up_read(&mdsc->snap_rwsem);
2081
2082         if (inode_dirty_flags)
2083                 __mark_inode_dirty(inode, inode_dirty_flags);
2084
2085
2086         if (mask) {
2087                 req->r_inode = inode;
2088                 ihold(inode);
2089                 req->r_inode_drop = release;
2090                 req->r_args.setattr.mask = cpu_to_le32(mask);
2091                 req->r_num_caps = 1;
2092                 req->r_stamp = attr->ia_ctime;
2093                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2094         }
2095         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2096              ceph_cap_string(dirtied), mask);
2097
2098         ceph_mdsc_put_request(req);
2099         ceph_free_cap_flush(prealloc_cf);
2100
2101         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2102                 __ceph_do_pending_vmtruncate(inode);
2103
2104         return err;
2105 }
2106
2107 /*
2108  * setattr
2109  */
2110 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2111 {
2112         struct inode *inode = d_inode(dentry);
2113         int err;
2114
2115         if (ceph_snap(inode) != CEPH_NOSNAP)
2116                 return -EROFS;
2117
2118         err = setattr_prepare(dentry, attr);
2119         if (err != 0)
2120                 return err;
2121
2122         err = __ceph_setattr(inode, attr);
2123
2124         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2125                 err = posix_acl_chmod(inode, attr->ia_mode);
2126
2127         return err;
2128 }
2129
2130 /*
2131  * Verify that we have a lease on the given mask.  If not,
2132  * do a getattr against an mds.
2133  */
2134 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2135                       int mask, bool force)
2136 {
2137         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2138         struct ceph_mds_client *mdsc = fsc->mdsc;
2139         struct ceph_mds_request *req;
2140         int err;
2141
2142         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2143                 dout("do_getattr inode %p SNAPDIR\n", inode);
2144                 return 0;
2145         }
2146
2147         dout("do_getattr inode %p mask %s mode 0%o\n",
2148              inode, ceph_cap_string(mask), inode->i_mode);
2149         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2150                 return 0;
2151
2152         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2153         if (IS_ERR(req))
2154                 return PTR_ERR(req);
2155         req->r_inode = inode;
2156         ihold(inode);
2157         req->r_num_caps = 1;
2158         req->r_args.getattr.mask = cpu_to_le32(mask);
2159         req->r_locked_page = locked_page;
2160         err = ceph_mdsc_do_request(mdsc, NULL, req);
2161         if (locked_page && err == 0) {
2162                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2163                 if (inline_version == 0) {
2164                         /* the reply is supposed to contain inline data */
2165                         err = -EINVAL;
2166                 } else if (inline_version == CEPH_INLINE_NONE) {
2167                         err = -ENODATA;
2168                 } else {
2169                         err = req->r_reply_info.targeti.inline_len;
2170                 }
2171         }
2172         ceph_mdsc_put_request(req);
2173         dout("do_getattr result=%d\n", err);
2174         return err;
2175 }
2176
2177
2178 /*
2179  * Check inode permissions.  We verify we have a valid value for
2180  * the AUTH cap, then call the generic handler.
2181  */
2182 int ceph_permission(struct inode *inode, int mask)
2183 {
2184         int err;
2185
2186         if (mask & MAY_NOT_BLOCK)
2187                 return -ECHILD;
2188
2189         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2190
2191         if (!err)
2192                 err = generic_permission(inode, mask);
2193         return err;
2194 }
2195
2196 /*
2197  * Get all attributes.  Hopefully somedata we'll have a statlite()
2198  * and can limit the fields we require to be accurate.
2199  */
2200 int ceph_getattr(const struct path *path, struct kstat *stat,
2201                  u32 request_mask, unsigned int flags)
2202 {
2203         struct inode *inode = d_inode(path->dentry);
2204         struct ceph_inode_info *ci = ceph_inode(inode);
2205         int err;
2206
2207         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2208         if (!err) {
2209                 generic_fillattr(inode, stat);
2210                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2211                 if (ceph_snap(inode) != CEPH_NOSNAP)
2212                         stat->dev = ceph_snap(inode);
2213                 else
2214                         stat->dev = 0;
2215                 if (S_ISDIR(inode->i_mode)) {
2216                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2217                                                 RBYTES))
2218                                 stat->size = ci->i_rbytes;
2219                         else
2220                                 stat->size = ci->i_files + ci->i_subdirs;
2221                         stat->blocks = 0;
2222                         stat->blksize = 65536;
2223                 }
2224         }
2225         return err;
2226 }