fs/ceph/inode.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/slab.h>
   6 #include <linux/string.h>
   7 #include <linux/uaccess.h>
   8 #include <linux/kernel.h>
   9 #include <linux/namei.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12
  13 #include "super.h"
  14 #include "mds_client.h"
  15 #include <linux/ceph/decode.h>
  16
  17 /*
  18  * Ceph inode operations
  19  *
  20  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  21  * setattr, etc.), xattr helpers, and helpers for assimilating
  22  * metadata returned by the MDS into our cache.
  23  *
  24  * Also define helpers for doing asynchronous writeback, invalidation,
  25  * and truncation for the benefit of those who can't afford to block
  26  * (typically because they are in the message handler path).
  27  */
  28
  29 static const struct inode_operations ceph_symlink_iops;
  30
  31 static void ceph_invalidate_work(struct work_struct *work);
  32 static void ceph_writeback_work(struct work_struct *work);
  33 static void ceph_vmtruncate_work(struct work_struct *work);
  34
  35 /*
  36  * find or create an inode, given the ceph ino number
  37  */
  38 static int ceph_set_ino_cb(struct inode *inode, void *data)
  39 {
  40         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  41         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  42         return 0;
  43 }
  44
  45 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  46 {
  47         struct inode *inode;
  48         ino_t t = ceph_vino_to_ino(vino);
  49
  50         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  51         if (inode == NULL)
  52                 return ERR_PTR(-ENOMEM);
  53         if (inode->i_state & I_NEW) {
  54                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  55                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  56                 unlock_new_inode(inode);
  57         }
  58
  59         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  60              vino.snap, inode);
  61         return inode;
  62 }
  63
  64 /*
  65  * get/constuct snapdir inode for a given directory
  66  */
  67 struct inode *ceph_get_snapdir(struct inode *parent)
  68 {
  69         struct ceph_vino vino = {
  70                 .ino = ceph_ino(parent),
  71                 .snap = CEPH_SNAPDIR,
  72         };
  73         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  74         struct ceph_inode_info *ci = ceph_inode(inode);
  75
  76         BUG_ON(!S_ISDIR(parent->i_mode));
  77         if (IS_ERR(inode))
  78                 return inode;
  79         inode->i_mode = parent->i_mode;
  80         inode->i_uid = parent->i_uid;
  81         inode->i_gid = parent->i_gid;
  82         inode->i_op = &ceph_dir_iops;
  83         inode->i_fop = &ceph_dir_fops;
  84         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  85         ci->i_rbytes = 0;
  86         return inode;
  87 }
  88
  89 const struct inode_operations ceph_file_iops = {
  90         .permission = ceph_permission,
  91         .setattr = ceph_setattr,
  92         .getattr = ceph_getattr,
  93         .setxattr = ceph_setxattr,
  94         .getxattr = ceph_getxattr,
  95         .listxattr = ceph_listxattr,
  96         .removexattr = ceph_removexattr,
  97 };
  98
  99
 100 /*
 101  * We use a 'frag tree' to keep track of the MDS's directory fragments
 102  * for a given inode (usually there is just a single fragment).  We
 103  * need to know when a child frag is delegated to a new MDS, or when
 104  * it is flagged as replicated, so we can direct our requests
 105  * accordingly.
 106  */
 107
 108 /*
 109  * find/create a frag in the tree
 110  */
 111 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 112                                                     u32 f)
 113 {
 114         struct rb_node **p;
 115         struct rb_node *parent = NULL;
 116         struct ceph_inode_frag *frag;
 117         int c;
 118
 119         p = &ci->i_fragtree.rb_node;
 120         while (*p) {
 121                 parent = *p;
 122                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 123                 c = ceph_frag_compare(f, frag->frag);
 124                 if (c < 0)
 125                         p = &(*p)->rb_left;
 126                 else if (c > 0)
 127                         p = &(*p)->rb_right;
 128                 else
 129                         return frag;
 130         }
 131
 132         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 133         if (!frag) {
 134                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 135                        "frag %x\n", &ci->vfs_inode,
 136                        ceph_vinop(&ci->vfs_inode), f);
 137                 return ERR_PTR(-ENOMEM);
 138         }
 139         frag->frag = f;
 140         frag->split_by = 0;
 141         frag->mds = -1;
 142         frag->ndist = 0;
 143
 144         rb_link_node(&frag->node, parent, p);
 145         rb_insert_color(&frag->node, &ci->i_fragtree);
 146
 147         dout("get_or_create_frag added %llx.%llx frag %x\n",
 148              ceph_vinop(&ci->vfs_inode), f);
 149         return frag;
 150 }
 151
 152 /*
 153  * find a specific frag @f
 154  */
 155 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 156 {
 157         struct rb_node *n = ci->i_fragtree.rb_node;
 158
 159         while (n) {
 160                 struct ceph_inode_frag *frag =
 161                         rb_entry(n, struct ceph_inode_frag, node);
 162                 int c = ceph_frag_compare(f, frag->frag);
 163                 if (c < 0)
 164                         n = n->rb_left;
 165                 else if (c > 0)
 166                         n = n->rb_right;
 167                 else
 168                         return frag;
 169         }
 170         return NULL;
 171 }
 172
 173 /*
 174  * Choose frag containing the given value @v.  If @pfrag is
 175  * specified, copy the frag delegation info to the caller if
 176  * it is present.
 177  */
 178 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 179                      struct ceph_inode_frag *pfrag,
 180                      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         mutex_lock(&ci->i_fragtree_mutex);
 191         while (1) {
 192                 WARN_ON(!ceph_frag_contains_value(t, v));
 193                 frag = __ceph_find_frag(ci, t);
 194                 if (!frag)
 195                         break; /* t is a leaf */
 196                 if (frag->split_by == 0) {
 197                         if (pfrag)
 198                                 memcpy(pfrag, frag, sizeof(*pfrag));
 199                         if (found)
 200                                 *found = 1;
 201                         break;
 202                 }
 203
 204                 /* choose child */
 205                 nway = 1 << frag->split_by;
 206                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 207                      frag->split_by, nway);
 208                 for (i = 0; i < nway; i++) {
 209                         n = ceph_frag_make_child(t, frag->split_by, i);
 210                         if (ceph_frag_contains_value(n, v)) {
 211                                 t = n;
 212                                 break;
 213                         }
 214                 }
 215                 BUG_ON(i == nway);
 216         }
 217         dout("choose_frag(%x) = %x\n", v, t);
 218
 219         mutex_unlock(&ci->i_fragtree_mutex);
 220         return t;
 221 }
 222
 223 /*
 224  * Process dirfrag (delegation) info from the mds.  Include leaf
 225  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 226  * branches/splits are included in i_fragtree)
 227  */
 228 static int ceph_fill_dirfrag(struct inode *inode,
 229                              struct ceph_mds_reply_dirfrag *dirinfo)
 230 {
 231         struct ceph_inode_info *ci = ceph_inode(inode);
 232         struct ceph_inode_frag *frag;
 233         u32 id = le32_to_cpu(dirinfo->frag);
 234         int mds = le32_to_cpu(dirinfo->auth);
 235         int ndist = le32_to_cpu(dirinfo->ndist);
 236         int i;
 237         int err = 0;
 238
 239         mutex_lock(&ci->i_fragtree_mutex);
 240         if (ndist == 0) {
 241                 /* no delegation info needed. */
 242                 frag = __ceph_find_frag(ci, id);
 243                 if (!frag)
 244                         goto out;
 245                 if (frag->split_by == 0) {
 246                         /* tree leaf, remove */
 247                         dout("fill_dirfrag removed %llx.%llx frag %x"
 248                              " (no ref)\n", ceph_vinop(inode), id);
 249                         rb_erase(&frag->node, &ci->i_fragtree);
 250                         kfree(frag);
 251                 } else {
 252                         /* tree branch, keep and clear */
 253                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 254                              " referral\n", ceph_vinop(inode), id);
 255                         frag->mds = -1;
 256                         frag->ndist = 0;
 257                 }
 258                 goto out;
 259         }
 260
 261
 262         /* find/add this frag to store mds delegation info */
 263         frag = __get_or_create_frag(ci, id);
 264         if (IS_ERR(frag)) {
 265                 /* this is not the end of the world; we can continue
 266                    with bad/inaccurate delegation info */
 267                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 268                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 269                 err = -ENOMEM;
 270                 goto out;
 271         }
 272
 273         frag->mds = mds;
 274         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 275         for (i = 0; i < frag->ndist; i++)
 276                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 277         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 278              ceph_vinop(inode), frag->frag, frag->ndist);
 279
 280 out:
 281         mutex_unlock(&ci->i_fragtree_mutex);
 282         return err;
 283 }
 284
 285
 286 /*
 287  * initialize a newly allocated inode.
 288  */
 289 struct inode *ceph_alloc_inode(struct super_block *sb)
 290 {
 291         struct ceph_inode_info *ci;
 292         int i;
 293
 294         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 295         if (!ci)
 296                 return NULL;
 297
 298         dout("alloc_inode %p\n", &ci->vfs_inode);
 299
 300         spin_lock_init(&ci->i_ceph_lock);
 301
 302         ci->i_version = 0;
 303         ci->i_time_warp_seq = 0;
 304         ci->i_ceph_flags = 0;
 305         atomic_set(&ci->i_release_count, 1);
 306         atomic_set(&ci->i_complete_count, 0);
 307         ci->i_symlink = NULL;
 308
 309         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 310
 311         ci->i_fragtree = RB_ROOT;
 312         mutex_init(&ci->i_fragtree_mutex);
 313
 314         ci->i_xattrs.blob = NULL;
 315         ci->i_xattrs.prealloc_blob = NULL;
 316         ci->i_xattrs.dirty = false;
 317         ci->i_xattrs.index = RB_ROOT;
 318         ci->i_xattrs.count = 0;
 319         ci->i_xattrs.names_size = 0;
 320         ci->i_xattrs.vals_size = 0;
 321         ci->i_xattrs.version = 0;
 322         ci->i_xattrs.index_version = 0;
 323
 324         ci->i_caps = RB_ROOT;
 325         ci->i_auth_cap = NULL;
 326         ci->i_dirty_caps = 0;
 327         ci->i_flushing_caps = 0;
 328         INIT_LIST_HEAD(&ci->i_dirty_item);
 329         INIT_LIST_HEAD(&ci->i_flushing_item);
 330         ci->i_cap_flush_seq = 0;
 331         ci->i_cap_flush_last_tid = 0;
 332         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 333         init_waitqueue_head(&ci->i_cap_wq);
 334         ci->i_hold_caps_min = 0;
 335         ci->i_hold_caps_max = 0;
 336         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 337         ci->i_cap_exporting_mds = 0;
 338         ci->i_cap_exporting_mseq = 0;
 339         ci->i_cap_exporting_issued = 0;
 340         INIT_LIST_HEAD(&ci->i_cap_snaps);
 341         ci->i_head_snapc = NULL;
 342         ci->i_snap_caps = 0;
 343
 344         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 345                 ci->i_nr_by_mode[i] = 0;
 346
 347         ci->i_truncate_seq = 0;
 348         ci->i_truncate_size = 0;
 349         ci->i_truncate_pending = 0;
 350
 351         ci->i_max_size = 0;
 352         ci->i_reported_size = 0;
 353         ci->i_wanted_max_size = 0;
 354         ci->i_requested_max_size = 0;
 355
 356         ci->i_pin_ref = 0;
 357         ci->i_rd_ref = 0;
 358         ci->i_rdcache_ref = 0;
 359         ci->i_wr_ref = 0;
 360         ci->i_wb_ref = 0;
 361         ci->i_wrbuffer_ref = 0;
 362         ci->i_wrbuffer_ref_head = 0;
 363         ci->i_shared_gen = 0;
 364         ci->i_rdcache_gen = 0;
 365         ci->i_rdcache_revoking = 0;
 366
 367         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 368         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 369         spin_lock_init(&ci->i_unsafe_lock);
 370
 371         ci->i_snap_realm = NULL;
 372         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 373         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 374
 375         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 376         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 377
 378         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 379
 380         return &ci->vfs_inode;
 381 }
 382
 383 static void ceph_i_callback(struct rcu_head *head)
 384 {
 385         struct inode *inode = container_of(head, struct inode, i_rcu);
 386         struct ceph_inode_info *ci = ceph_inode(inode);
 387
 388         kmem_cache_free(ceph_inode_cachep, ci);
 389 }
 390
 391 void ceph_destroy_inode(struct inode *inode)
 392 {
 393         struct ceph_inode_info *ci = ceph_inode(inode);
 394         struct ceph_inode_frag *frag;
 395         struct rb_node *n;
 396
 397         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 398
 399         ceph_queue_caps_release(inode);
 400
 401         /*
 402          * we may still have a snap_realm reference if there are stray
 403          * caps in i_cap_exporting_issued or i_snap_caps.
 404          */
 405         if (ci->i_snap_realm) {
 406                 struct ceph_mds_client *mdsc =
 407                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 408                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 409
 410                 dout(" dropping residual ref to snap realm %p\n", realm);
 411                 spin_lock(&realm->inodes_with_caps_lock);
 412                 list_del_init(&ci->i_snap_realm_item);
 413                 spin_unlock(&realm->inodes_with_caps_lock);
 414                 ceph_put_snap_realm(mdsc, realm);
 415         }
 416
 417         kfree(ci->i_symlink);
 418         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 419                 frag = rb_entry(n, struct ceph_inode_frag, node);
 420                 rb_erase(n, &ci->i_fragtree);
 421                 kfree(frag);
 422         }
 423
 424         __ceph_destroy_xattrs(ci);
 425         if (ci->i_xattrs.blob)
 426                 ceph_buffer_put(ci->i_xattrs.blob);
 427         if (ci->i_xattrs.prealloc_blob)
 428                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 429
 430         call_rcu(&inode->i_rcu, ceph_i_callback);
 431 }
 432
 433
 434 /*
 435  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 436  * careful because either the client or MDS may have more up to date
 437  * info, depending on which capabilities are held, and whether
 438  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 439  * and size are monotonically increasing, except when utimes() or
 440  * truncate() increments the corresponding _seq values.)
 441  */
 442 int ceph_fill_file_size(struct inode *inode, int issued,
 443                         u32 truncate_seq, u64 truncate_size, u64 size)
 444 {
 445         struct ceph_inode_info *ci = ceph_inode(inode);
 446         int queue_trunc = 0;
 447
 448         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 449             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 450                 dout("size %lld -> %llu\n", inode->i_size, size);
 451                 inode->i_size = size;
 452                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 453                 ci->i_reported_size = size;
 454                 if (truncate_seq != ci->i_truncate_seq) {
 455                         dout("truncate_seq %u -> %u\n",
 456                              ci->i_truncate_seq, truncate_seq);
 457                         ci->i_truncate_seq = truncate_seq;
 458                         /*
 459                          * If we hold relevant caps, or in the case where we're
 460                          * not the only client referencing this file and we
 461                          * don't hold those caps, then we need to check whether
 462                          * the file is either opened or mmaped
 463                          */
 464                         if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
 465                                        CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
 466                                        CEPH_CAP_FILE_EXCL|
 467                                        CEPH_CAP_FILE_LAZYIO)) ||
 468                             mapping_mapped(inode->i_mapping) ||
 469                             __ceph_caps_file_wanted(ci)) {
 470                                 ci->i_truncate_pending++;
 471                                 queue_trunc = 1;
 472                         }
 473                 }
 474         }
 475         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 476             ci->i_truncate_size != truncate_size) {
 477                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 478                      truncate_size);
 479                 ci->i_truncate_size = truncate_size;
 480         }
 481         return queue_trunc;
 482 }
 483
 484 void ceph_fill_file_time(struct inode *inode, int issued,
 485                          u64 time_warp_seq, struct timespec *ctime,
 486                          struct timespec *mtime, struct timespec *atime)
 487 {
 488         struct ceph_inode_info *ci = ceph_inode(inode);
 489         int warn = 0;
 490
 491         if (issued & (CEPH_CAP_FILE_EXCL|
 492                       CEPH_CAP_FILE_WR|
 493                       CEPH_CAP_FILE_BUFFER|
 494                       CEPH_CAP_AUTH_EXCL|
 495                       CEPH_CAP_XATTR_EXCL)) {
 496                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 497                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 498                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 499                              ctime->tv_sec, ctime->tv_nsec);
 500                         inode->i_ctime = *ctime;
 501                 }
 502                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 503                         /* the MDS did a utimes() */
 504                         dout("mtime %ld.%09ld -> %ld.%09ld "
 505                              "tw %d -> %d\n",
 506                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 507                              mtime->tv_sec, mtime->tv_nsec,
 508                              ci->i_time_warp_seq, (int)time_warp_seq);
 509
 510                         inode->i_mtime = *mtime;
 511                         inode->i_atime = *atime;
 512                         ci->i_time_warp_seq = time_warp_seq;
 513                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 514                         /* nobody did utimes(); take the max */
 515                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 516                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 517                                      inode->i_mtime.tv_sec,
 518                                      inode->i_mtime.tv_nsec,
 519                                      mtime->tv_sec, mtime->tv_nsec);
 520                                 inode->i_mtime = *mtime;
 521                         }
 522                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 523                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 524                                      inode->i_atime.tv_sec,
 525                                      inode->i_atime.tv_nsec,
 526                                      atime->tv_sec, atime->tv_nsec);
 527                                 inode->i_atime = *atime;
 528                         }
 529                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 530                         /* we did a utimes(); ignore mds values */
 531                 } else {
 532                         warn = 1;
 533                 }
 534         } else {
 535                 /* we have no write|excl caps; whatever the MDS says is true */
 536                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 537                         inode->i_ctime = *ctime;
 538                         inode->i_mtime = *mtime;
 539                         inode->i_atime = *atime;
 540                         ci->i_time_warp_seq = time_warp_seq;
 541                 } else {
 542                         warn = 1;
 543                 }
 544         }
 545         if (warn) /* time_warp_seq shouldn't go backwards */
 546                 dout("%p mds time_warp_seq %llu < %u\n",
 547                      inode, time_warp_seq, ci->i_time_warp_seq);
 548 }
 549
 550 /*
 551  * Populate an inode based on info from mds.  May be called on new or
 552  * existing inodes.
 553  */
 554 static int fill_inode(struct inode *inode,
 555                       struct ceph_mds_reply_info_in *iinfo,
 556                       struct ceph_mds_reply_dirfrag *dirinfo,
 557                       struct ceph_mds_session *session,
 558                       unsigned long ttl_from, int cap_fmode,
 559                       struct ceph_cap_reservation *caps_reservation)
 560 {
 561         struct ceph_mds_reply_inode *info = iinfo->in;
 562         struct ceph_inode_info *ci = ceph_inode(inode);
 563         int i;
 564         int issued = 0, implemented;
 565         struct timespec mtime, atime, ctime;
 566         u32 nsplits;
 567         struct ceph_buffer *xattr_blob = NULL;
 568         int err = 0;
 569         int queue_trunc = 0;
 570
 571         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 572              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 573              ci->i_version);
 574
 575         /*
 576          * prealloc xattr data, if it looks like we'll need it.  only
 577          * if len > 4 (meaning there are actually xattrs; the first 4
 578          * bytes are the xattr count).
 579          */
 580         if (iinfo->xattr_len > 4) {
 581                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 582                 if (!xattr_blob)
 583                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 584                                iinfo->xattr_len);
 585         }
 586
 587         spin_lock(&ci->i_ceph_lock);
 588
 589         /*
 590          * provided version will be odd if inode value is projected,
 591          * even if stable.  skip the update if we have newer stable
 592          * info (ours>=theirs, e.g. due to racing mds replies), unless
 593          * we are getting projected (unstable) info (in which case the
 594          * version is odd, and we want ours>theirs).
 595          *   us   them
 596          *   2    2     skip
 597          *   3    2     skip
 598          *   3    3     update
 599          */
 600         if (le64_to_cpu(info->version) > 0 &&
 601             (ci->i_version & ~1) >= le64_to_cpu(info->version))
 602                 goto no_change;
 603
 604         issued = __ceph_caps_issued(ci, &implemented);
 605         issued |= implemented | __ceph_caps_dirty(ci);
 606
 607         /* update inode */
 608         ci->i_version = le64_to_cpu(info->version);
 609         inode->i_version++;
 610         inode->i_rdev = le32_to_cpu(info->rdev);
 611
 612         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
 613                 inode->i_mode = le32_to_cpu(info->mode);
 614                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 615                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 616                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 617                      from_kuid(&init_user_ns, inode->i_uid),
 618                      from_kgid(&init_user_ns, inode->i_gid));
 619         }
 620
 621         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
 622                 set_nlink(inode, le32_to_cpu(info->nlink));
 623
 624         /* be careful with mtime, atime, size */
 625         ceph_decode_timespec(&atime, &info->atime);
 626         ceph_decode_timespec(&mtime, &info->mtime);
 627         ceph_decode_timespec(&ctime, &info->ctime);
 628         queue_trunc = ceph_fill_file_size(inode, issued,
 629                                           le32_to_cpu(info->truncate_seq),
 630                                           le64_to_cpu(info->truncate_size),
 631                                           le64_to_cpu(info->size));
 632         ceph_fill_file_time(inode, issued,
 633                             le32_to_cpu(info->time_warp_seq),
 634                             &ctime, &mtime, &atime);
 635
 636         /* only update max_size on auth cap */
 637         if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 638             ci->i_max_size != le64_to_cpu(info->max_size)) {
 639                 dout("max_size %lld -> %llu\n", ci->i_max_size,
 640                      le64_to_cpu(info->max_size));
 641                 ci->i_max_size = le64_to_cpu(info->max_size);
 642         }
 643
 644         ci->i_layout = info->layout;
 645         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 646
 647         /* xattrs */
 648         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 649         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
 650             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 651                 if (ci->i_xattrs.blob)
 652                         ceph_buffer_put(ci->i_xattrs.blob);
 653                 ci->i_xattrs.blob = xattr_blob;
 654                 if (xattr_blob)
 655                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 656                                iinfo->xattr_data, iinfo->xattr_len);
 657                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 658                 xattr_blob = NULL;
 659         }
 660
 661         inode->i_mapping->a_ops = &ceph_aops;
 662         inode->i_mapping->backing_dev_info =
 663                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
 664
 665         switch (inode->i_mode & S_IFMT) {
 666         case S_IFIFO:
 667         case S_IFBLK:
 668         case S_IFCHR:
 669         case S_IFSOCK:
 670                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 671                 inode->i_op = &ceph_file_iops;
 672                 break;
 673         case S_IFREG:
 674                 inode->i_op = &ceph_file_iops;
 675                 inode->i_fop = &ceph_file_fops;
 676                 break;
 677         case S_IFLNK:
 678                 inode->i_op = &ceph_symlink_iops;
 679                 if (!ci->i_symlink) {
 680                         u32 symlen = iinfo->symlink_len;
 681                         char *sym;
 682
 683                         spin_unlock(&ci->i_ceph_lock);
 684
 685                         err = -EINVAL;
 686                         if (WARN_ON(symlen != inode->i_size))
 687                                 goto out;
 688
 689                         err = -ENOMEM;
 690                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 691                         if (!sym)
 692                                 goto out;
 693
 694                         spin_lock(&ci->i_ceph_lock);
 695                         if (!ci->i_symlink)
 696                                 ci->i_symlink = sym;
 697                         else
 698                                 kfree(sym); /* lost a race */
 699                 }
 700                 break;
 701         case S_IFDIR:
 702                 inode->i_op = &ceph_dir_iops;
 703                 inode->i_fop = &ceph_dir_fops;
 704
 705                 ci->i_dir_layout = iinfo->dir_layout;
 706
 707                 ci->i_files = le64_to_cpu(info->files);
 708                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 709                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 710                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 711                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 712                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 713                 break;
 714         default:
 715                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 716                        ceph_vinop(inode), inode->i_mode);
 717         }
 718
 719         /* set dir completion flag? */
 720         if (S_ISDIR(inode->i_mode) &&
 721             ci->i_files == 0 && ci->i_subdirs == 0 &&
 722             ceph_snap(inode) == CEPH_NOSNAP &&
 723             (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
 724             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 725             !__ceph_dir_is_complete(ci)) {
 726                 dout(" marking %p complete (empty)\n", inode);
 727                 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
 728                 ci->i_max_offset = 2;
 729         }
 730 no_change:
 731         spin_unlock(&ci->i_ceph_lock);
 732
 733         /* queue truncate if we saw i_size decrease */
 734         if (queue_trunc)
 735                 ceph_queue_vmtruncate(inode);
 736
 737         /* populate frag tree */
 738         /* FIXME: move me up, if/when version reflects fragtree changes */
 739         nsplits = le32_to_cpu(info->fragtree.nsplits);
 740         mutex_lock(&ci->i_fragtree_mutex);
 741         for (i = 0; i < nsplits; i++) {
 742                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
 743                 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
 744
 745                 if (IS_ERR(frag))
 746                         continue;
 747                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
 748                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 749         }
 750         mutex_unlock(&ci->i_fragtree_mutex);
 751
 752         /* were we issued a capability? */
 753         if (info->cap.caps) {
 754                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 755                         ceph_add_cap(inode, session,
 756                                      le64_to_cpu(info->cap.cap_id),
 757                                      cap_fmode,
 758                                      le32_to_cpu(info->cap.caps),
 759                                      le32_to_cpu(info->cap.wanted),
 760                                      le32_to_cpu(info->cap.seq),
 761                                      le32_to_cpu(info->cap.mseq),
 762                                      le64_to_cpu(info->cap.realm),
 763                                      info->cap.flags,
 764                                      caps_reservation);
 765                 } else {
 766                         spin_lock(&ci->i_ceph_lock);
 767                         dout(" %p got snap_caps %s\n", inode,
 768                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 769                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 770                         if (cap_fmode >= 0)
 771                                 __ceph_get_fmode(ci, cap_fmode);
 772                         spin_unlock(&ci->i_ceph_lock);
 773                 }
 774         } else if (cap_fmode >= 0) {
 775                 pr_warning("mds issued no caps on %llx.%llx\n",
 776                            ceph_vinop(inode));
 777                 __ceph_get_fmode(ci, cap_fmode);
 778         }
 779
 780         /* update delegation info? */
 781         if (dirinfo)
 782                 ceph_fill_dirfrag(inode, dirinfo);
 783
 784         err = 0;
 785
 786 out:
 787         if (xattr_blob)
 788                 ceph_buffer_put(xattr_blob);
 789         return err;
 790 }
 791
 792 /*
 793  * caller should hold session s_mutex.
 794  */
 795 static void update_dentry_lease(struct dentry *dentry,
 796                                 struct ceph_mds_reply_lease *lease,
 797                                 struct ceph_mds_session *session,
 798                                 unsigned long from_time)
 799 {
 800         struct ceph_dentry_info *di = ceph_dentry(dentry);
 801         long unsigned duration = le32_to_cpu(lease->duration_ms);
 802         long unsigned ttl = from_time + (duration * HZ) / 1000;
 803         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 804         struct inode *dir;
 805
 806         /* only track leases on regular dentries */
 807         if (dentry->d_op != &ceph_dentry_ops)
 808                 return;
 809
 810         spin_lock(&dentry->d_lock);
 811         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 812              dentry, duration, ttl);
 813
 814         /* make lease_rdcache_gen match directory */
 815         dir = dentry->d_parent->d_inode;
 816         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 817
 818         if (duration == 0)
 819                 goto out_unlock;
 820
 821         if (di->lease_gen == session->s_cap_gen &&
 822             time_before(ttl, dentry->d_time))
 823                 goto out_unlock;  /* we already have a newer lease. */
 824
 825         if (di->lease_session && di->lease_session != session)
 826                 goto out_unlock;
 827
 828         ceph_dentry_lru_touch(dentry);
 829
 830         if (!di->lease_session)
 831                 di->lease_session = ceph_get_mds_session(session);
 832         di->lease_gen = session->s_cap_gen;
 833         di->lease_seq = le32_to_cpu(lease->seq);
 834         di->lease_renew_after = half_ttl;
 835         di->lease_renew_from = 0;
 836         dentry->d_time = ttl;
 837 out_unlock:
 838         spin_unlock(&dentry->d_lock);
 839         return;
 840 }
 841
 842 /*
 843  * Set dentry's directory position based on the current dir's max, and
 844  * order it in d_subdirs, so that dcache_readdir behaves.
 845  *
 846  * Always called under directory's i_mutex.
 847  */
 848 static void ceph_set_dentry_offset(struct dentry *dn)
 849 {
 850         struct dentry *dir = dn->d_parent;
 851         struct inode *inode = dir->d_inode;
 852         struct ceph_inode_info *ci;
 853         struct ceph_dentry_info *di;
 854
 855         BUG_ON(!inode);
 856
 857         ci = ceph_inode(inode);
 858         di = ceph_dentry(dn);
 859
 860         spin_lock(&ci->i_ceph_lock);
 861         if (!__ceph_dir_is_complete(ci)) {
 862                 spin_unlock(&ci->i_ceph_lock);
 863                 return;
 864         }
 865         di->offset = ceph_inode(inode)->i_max_offset++;
 866         spin_unlock(&ci->i_ceph_lock);
 867
 868         spin_lock(&dir->d_lock);
 869         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
 870         list_move(&dn->d_u.d_child, &dir->d_subdirs);
 871         dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
 872              dn->d_u.d_child.prev, dn->d_u.d_child.next);
 873         spin_unlock(&dn->d_lock);
 874         spin_unlock(&dir->d_lock);
 875 }
 876
 877 /*
 878  * splice a dentry to an inode.
 879  * caller must hold directory i_mutex for this to be safe.
 880  *
 881  * we will only rehash the resulting dentry if @prehash is
 882  * true; @prehash will be set to false (for the benefit of
 883  * the caller) if we fail.
 884  */
 885 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 886                                     bool *prehash, bool set_offset)
 887 {
 888         struct dentry *realdn;
 889
 890         BUG_ON(dn->d_inode);
 891
 892         /* dn must be unhashed */
 893         if (!d_unhashed(dn))
 894                 d_drop(dn);
 895         realdn = d_materialise_unique(dn, in);
 896         if (IS_ERR(realdn)) {
 897                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 898                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 899                 if (prehash)
 900                         *prehash = false; /* don't rehash on error */
 901                 dn = realdn; /* note realdn contains the error */
 902                 goto out;
 903         } else if (realdn) {
 904                 dout("dn %p (%d) spliced with %p (%d) "
 905                      "inode %p ino %llx.%llx\n",
 906                      dn, dn->d_count,
 907                      realdn, realdn->d_count,
 908                      realdn->d_inode, ceph_vinop(realdn->d_inode));
 909                 dput(dn);
 910                 dn = realdn;
 911         } else {
 912                 BUG_ON(!ceph_dentry(dn));
 913                 dout("dn %p attached to %p ino %llx.%llx\n",
 914                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
 915         }
 916         if ((!prehash || *prehash) && d_unhashed(dn))
 917                 d_rehash(dn);
 918         if (set_offset)
 919                 ceph_set_dentry_offset(dn);
 920 out:
 921         return dn;
 922 }
 923
 924 /*
 925  * Incorporate results into the local cache.  This is either just
 926  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
 927  * after a lookup).
 928  *
 929  * A reply may contain
 930  *         a directory inode along with a dentry.
 931  *  and/or a target inode
 932  *
 933  * Called with snap_rwsem (read).
 934  */
 935 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
 936                     struct ceph_mds_session *session)
 937 {
 938         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
 939         struct inode *in = NULL;
 940         struct ceph_mds_reply_inode *ininfo;
 941         struct ceph_vino vino;
 942         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 943         int i = 0;
 944         int err = 0;
 945
 946         dout("fill_trace %p is_dentry %d is_target %d\n", req,
 947              rinfo->head->is_dentry, rinfo->head->is_target);
 948
 949 #if 0
 950         /*
 951          * Debugging hook:
 952          *
 953          * If we resend completed ops to a recovering mds, we get no
 954          * trace.  Since that is very rare, pretend this is the case
 955          * to ensure the 'no trace' handlers in the callers behave.
 956          *
 957          * Fill in inodes unconditionally to avoid breaking cap
 958          * invariants.
 959          */
 960         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
 961                 pr_info("fill_trace faking empty trace on %lld %s\n",
 962                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
 963                 if (rinfo->head->is_dentry) {
 964                         rinfo->head->is_dentry = 0;
 965                         err = fill_inode(req->r_locked_dir,
 966                                          &rinfo->diri, rinfo->dirfrag,
 967                                          session, req->r_request_started, -1);
 968                 }
 969                 if (rinfo->head->is_target) {
 970                         rinfo->head->is_target = 0;
 971                         ininfo = rinfo->targeti.in;
 972                         vino.ino = le64_to_cpu(ininfo->ino);
 973                         vino.snap = le64_to_cpu(ininfo->snapid);
 974                         in = ceph_get_inode(sb, vino);
 975                         err = fill_inode(in, &rinfo->targeti, NULL,
 976                                          session, req->r_request_started,
 977                                          req->r_fmode);
 978                         iput(in);
 979                 }
 980         }
 981 #endif
 982
 983         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
 984                 dout("fill_trace reply is empty!\n");
 985                 if (rinfo->head->result == 0 && req->r_locked_dir)
 986                         ceph_invalidate_dir_request(req);
 987                 return 0;
 988         }
 989
 990         if (rinfo->head->is_dentry) {
 991                 struct inode *dir = req->r_locked_dir;
 992
 993                 if (dir) {
 994                         err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
 995                                          session, req->r_request_started, -1,
 996                                          &req->r_caps_reservation);
 997                         if (err < 0)
 998                                 return err;
 999                 } else {
1000                         WARN_ON_ONCE(1);
1001                 }
1002         }
1003
1004         /*
1005          * ignore null lease/binding on snapdir ENOENT, or else we
1006          * will have trouble splicing in the virtual snapdir later
1007          */
1008         if (rinfo->head->is_dentry && !req->r_aborted &&
1009             req->r_locked_dir &&
1010             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1011                                                fsc->mount_options->snapdir_name,
1012                                                req->r_dentry->d_name.len))) {
1013                 /*
1014                  * lookup link rename   : null -> possibly existing inode
1015                  * mknod symlink mkdir  : null -> new inode
1016                  * unlink               : linked -> null
1017                  */
1018                 struct inode *dir = req->r_locked_dir;
1019                 struct dentry *dn = req->r_dentry;
1020                 bool have_dir_cap, have_lease;
1021
1022                 BUG_ON(!dn);
1023                 BUG_ON(!dir);
1024                 BUG_ON(dn->d_parent->d_inode != dir);
1025                 BUG_ON(ceph_ino(dir) !=
1026                        le64_to_cpu(rinfo->diri.in->ino));
1027                 BUG_ON(ceph_snap(dir) !=
1028                        le64_to_cpu(rinfo->diri.in->snapid));
1029
1030                 /* do we have a lease on the whole dir? */
1031                 have_dir_cap =
1032                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1033                          CEPH_CAP_FILE_SHARED);
1034
1035                 /* do we have a dn lease? */
1036                 have_lease = have_dir_cap ||
1037                         le32_to_cpu(rinfo->dlease->duration_ms);
1038                 if (!have_lease)
1039                         dout("fill_trace  no dentry lease or dir cap\n");
1040
1041                 /* rename? */
1042                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1043                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1044                              req->r_old_dentry,
1045                              req->r_old_dentry->d_name.len,
1046                              req->r_old_dentry->d_name.name,
1047                              dn, dn->d_name.len, dn->d_name.name);
1048                         dout("fill_trace doing d_move %p -> %p\n",
1049                              req->r_old_dentry, dn);
1050
1051                         d_move(req->r_old_dentry, dn);
1052                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1053                              req->r_old_dentry,
1054                              req->r_old_dentry->d_name.len,
1055                              req->r_old_dentry->d_name.name,
1056                              dn, dn->d_name.len, dn->d_name.name);
1057
1058                         /* ensure target dentry is invalidated, despite
1059                            rehashing bug in vfs_rename_dir */
1060                         ceph_invalidate_dentry_lease(dn);
1061
1062                         /*
1063                          * d_move() puts the renamed dentry at the end of
1064                          * d_subdirs.  We need to assign it an appropriate
1065                          * directory offset so we can behave when dir is
1066                          * complete.
1067                          */
1068                         ceph_set_dentry_offset(req->r_old_dentry);
1069                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1070                              ceph_dentry(req->r_old_dentry)->offset);
1071
1072                         dn = req->r_old_dentry;  /* use old_dentry */
1073                         in = dn->d_inode;
1074                 }
1075
1076                 /* null dentry? */
1077                 if (!rinfo->head->is_target) {
1078                         dout("fill_trace null dentry\n");
1079                         if (dn->d_inode) {
1080                                 dout("d_delete %p\n", dn);
1081                                 d_delete(dn);
1082                         } else {
1083                                 dout("d_instantiate %p NULL\n", dn);
1084                                 d_instantiate(dn, NULL);
1085                                 if (have_lease && d_unhashed(dn))
1086                                         d_rehash(dn);
1087                                 update_dentry_lease(dn, rinfo->dlease,
1088                                                     session,
1089                                                     req->r_request_started);
1090                         }
1091                         goto done;
1092                 }
1093
1094                 /* attach proper inode */
1095                 ininfo = rinfo->targeti.in;
1096                 vino.ino = le64_to_cpu(ininfo->ino);
1097                 vino.snap = le64_to_cpu(ininfo->snapid);
1098                 in = dn->d_inode;
1099                 if (!in) {
1100                         in = ceph_get_inode(sb, vino);
1101                         if (IS_ERR(in)) {
1102                                 pr_err("fill_trace bad get_inode "
1103                                        "%llx.%llx\n", vino.ino, vino.snap);
1104                                 err = PTR_ERR(in);
1105                                 d_drop(dn);
1106                                 goto done;
1107                         }
1108                         dn = splice_dentry(dn, in, &have_lease, true);
1109                         if (IS_ERR(dn)) {
1110                                 err = PTR_ERR(dn);
1111                                 goto done;
1112                         }
1113                         req->r_dentry = dn;  /* may have spliced */
1114                         ihold(in);
1115                 } else if (ceph_ino(in) == vino.ino &&
1116                            ceph_snap(in) == vino.snap) {
1117                         ihold(in);
1118                 } else {
1119                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1120                              dn, in, ceph_ino(in), ceph_snap(in),
1121                              vino.ino, vino.snap);
1122                         have_lease = false;
1123                         in = NULL;
1124                 }
1125
1126                 if (have_lease)
1127                         update_dentry_lease(dn, rinfo->dlease, session,
1128                                             req->r_request_started);
1129                 dout(" final dn %p\n", dn);
1130                 i++;
1131         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1132                    req->r_op == CEPH_MDS_OP_MKSNAP) && !req->r_aborted) {
1133                 struct dentry *dn = req->r_dentry;
1134
1135                 /* fill out a snapdir LOOKUPSNAP dentry */
1136                 BUG_ON(!dn);
1137                 BUG_ON(!req->r_locked_dir);
1138                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1139                 ininfo = rinfo->targeti.in;
1140                 vino.ino = le64_to_cpu(ininfo->ino);
1141                 vino.snap = le64_to_cpu(ininfo->snapid);
1142                 in = ceph_get_inode(sb, vino);
1143                 if (IS_ERR(in)) {
1144                         pr_err("fill_inode get_inode badness %llx.%llx\n",
1145                                vino.ino, vino.snap);
1146                         err = PTR_ERR(in);
1147                         d_delete(dn);
1148                         goto done;
1149                 }
1150                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1151                 dn = splice_dentry(dn, in, NULL, true);
1152                 if (IS_ERR(dn)) {
1153                         err = PTR_ERR(dn);
1154                         goto done;
1155                 }
1156                 req->r_dentry = dn;  /* may have spliced */
1157                 ihold(in);
1158                 rinfo->head->is_dentry = 1;  /* fool notrace handlers */
1159         }
1160
1161         if (rinfo->head->is_target) {
1162                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1163                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1164
1165                 if (in == NULL || ceph_ino(in) != vino.ino ||
1166                     ceph_snap(in) != vino.snap) {
1167                         in = ceph_get_inode(sb, vino);
1168                         if (IS_ERR(in)) {
1169                                 err = PTR_ERR(in);
1170                                 goto done;
1171                         }
1172                 }
1173                 req->r_target_inode = in;
1174
1175                 err = fill_inode(in,
1176                                  &rinfo->targeti, NULL,
1177                                  session, req->r_request_started,
1178                                  (le32_to_cpu(rinfo->head->result) == 0) ?
1179                                  req->r_fmode : -1,
1180                                  &req->r_caps_reservation);
1181                 if (err < 0) {
1182                         pr_err("fill_inode badness %p %llx.%llx\n",
1183                                in, ceph_vinop(in));
1184                         goto done;
1185                 }
1186         }
1187
1188 done:
1189         dout("fill_trace done err=%d\n", err);
1190         return err;
1191 }
1192
1193 /*
1194  * Prepopulate our cache with readdir results, leases, etc.
1195  */
1196 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1197                                            struct ceph_mds_session *session)
1198 {
1199         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1200         int i, err = 0;
1201
1202         for (i = 0; i < rinfo->dir_nr; i++) {
1203                 struct ceph_vino vino;
1204                 struct inode *in;
1205                 int rc;
1206
1207                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1208                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1209
1210                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1211                 if (IS_ERR(in)) {
1212                         err = PTR_ERR(in);
1213                         dout("new_inode badness got %d\n", err);
1214                         continue;
1215                 }
1216                 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1217                                 req->r_request_started, -1,
1218                                 &req->r_caps_reservation);
1219                 if (rc < 0) {
1220                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1221                         err = rc;
1222                         continue;
1223                 }
1224         }
1225
1226         return err;
1227 }
1228
1229 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1230                              struct ceph_mds_session *session)
1231 {
1232         struct dentry *parent = req->r_dentry;
1233         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1234         struct qstr dname;
1235         struct dentry *dn;
1236         struct inode *in;
1237         int err = 0, i;
1238         struct inode *snapdir = NULL;
1239         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1240         u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1241         struct ceph_dentry_info *di;
1242
1243         if (req->r_aborted)
1244                 return readdir_prepopulate_inodes_only(req, session);
1245
1246         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1247                 snapdir = ceph_get_snapdir(parent->d_inode);
1248                 parent = d_find_alias(snapdir);
1249                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1250                      rinfo->dir_nr, parent);
1251         } else {
1252                 dout("readdir_prepopulate %d items under dn %p\n",
1253                      rinfo->dir_nr, parent);
1254                 if (rinfo->dir_dir)
1255                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1256         }
1257
1258         for (i = 0; i < rinfo->dir_nr; i++) {
1259                 struct ceph_vino vino;
1260
1261                 dname.name = rinfo->dir_dname[i];
1262                 dname.len = rinfo->dir_dname_len[i];
1263                 dname.hash = full_name_hash(dname.name, dname.len);
1264
1265                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1266                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1267
1268 retry_lookup:
1269                 dn = d_lookup(parent, &dname);
1270                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1271                      parent, dname.len, dname.name, dn);
1272
1273                 if (!dn) {
1274                         dn = d_alloc(parent, &dname);
1275                         dout("d_alloc %p '%.*s' = %p\n", parent,
1276                              dname.len, dname.name, dn);
1277                         if (dn == NULL) {
1278                                 dout("d_alloc badness\n");
1279                                 err = -ENOMEM;
1280                                 goto out;
1281                         }
1282                         err = ceph_init_dentry(dn);
1283                         if (err < 0) {
1284                                 dput(dn);
1285                                 goto out;
1286                         }
1287                 } else if (dn->d_inode &&
1288                            (ceph_ino(dn->d_inode) != vino.ino ||
1289                             ceph_snap(dn->d_inode) != vino.snap)) {
1290                         dout(" dn %p points to wrong inode %p\n",
1291                              dn, dn->d_inode);
1292                         d_delete(dn);
1293                         dput(dn);
1294                         goto retry_lookup;
1295                 } else {
1296                         /* reorder parent's d_subdirs */
1297                         spin_lock(&parent->d_lock);
1298                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1299                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1300                         spin_unlock(&dn->d_lock);
1301                         spin_unlock(&parent->d_lock);
1302                 }
1303
1304                 di = dn->d_fsdata;
1305                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1306
1307                 /* inode */
1308                 if (dn->d_inode) {
1309                         in = dn->d_inode;
1310                 } else {
1311                         in = ceph_get_inode(parent->d_sb, vino);
1312                         if (IS_ERR(in)) {
1313                                 dout("new_inode badness\n");
1314                                 d_drop(dn);
1315                                 dput(dn);
1316                                 err = PTR_ERR(in);
1317                                 goto out;
1318                         }
1319                         dn = splice_dentry(dn, in, NULL, false);
1320                         if (IS_ERR(dn))
1321                                 dn = NULL;
1322                 }
1323
1324                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1325                                req->r_request_started, -1,
1326                                &req->r_caps_reservation) < 0) {
1327                         pr_err("fill_inode badness on %p\n", in);
1328                         goto next_item;
1329                 }
1330                 if (dn)
1331                         update_dentry_lease(dn, rinfo->dir_dlease[i],
1332                                             req->r_session,
1333                                             req->r_request_started);
1334 next_item:
1335                 if (dn)
1336                         dput(dn);
1337         }
1338         req->r_did_prepopulate = true;
1339
1340 out:
1341         if (snapdir) {
1342                 iput(snapdir);
1343                 dput(parent);
1344         }
1345         dout("readdir_prepopulate done\n");
1346         return err;
1347 }
1348
1349 int ceph_inode_set_size(struct inode *inode, loff_t size)
1350 {
1351         struct ceph_inode_info *ci = ceph_inode(inode);
1352         int ret = 0;
1353
1354         spin_lock(&ci->i_ceph_lock);
1355         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1356         inode->i_size = size;
1357         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1358
1359         /* tell the MDS if we are approaching max_size */
1360         if ((size << 1) >= ci->i_max_size &&
1361             (ci->i_reported_size << 1) < ci->i_max_size)
1362                 ret = 1;
1363
1364         spin_unlock(&ci->i_ceph_lock);
1365         return ret;
1366 }
1367
1368 /*
1369  * Write back inode data in a worker thread.  (This can't be done
1370  * in the message handler context.)
1371  */
1372 void ceph_queue_writeback(struct inode *inode)
1373 {
1374         ihold(inode);
1375         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1376                        &ceph_inode(inode)->i_wb_work)) {
1377                 dout("ceph_queue_writeback %p\n", inode);
1378         } else {
1379                 dout("ceph_queue_writeback %p failed\n", inode);
1380                 iput(inode);
1381         }
1382 }
1383
1384 static void ceph_writeback_work(struct work_struct *work)
1385 {
1386         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1387                                                   i_wb_work);
1388         struct inode *inode = &ci->vfs_inode;
1389
1390         dout("writeback %p\n", inode);
1391         filemap_fdatawrite(&inode->i_data);
1392         iput(inode);
1393 }
1394
1395 /*
1396  * queue an async invalidation
1397  */
1398 void ceph_queue_invalidate(struct inode *inode)
1399 {
1400         ihold(inode);
1401         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1402                        &ceph_inode(inode)->i_pg_inv_work)) {
1403                 dout("ceph_queue_invalidate %p\n", inode);
1404         } else {
1405                 dout("ceph_queue_invalidate %p failed\n", inode);
1406                 iput(inode);
1407         }
1408 }
1409
1410 /*
1411  * Invalidate inode pages in a worker thread.  (This can't be done
1412  * in the message handler context.)
1413  */
1414 static void ceph_invalidate_work(struct work_struct *work)
1415 {
1416         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1417                                                   i_pg_inv_work);
1418         struct inode *inode = &ci->vfs_inode;
1419         u32 orig_gen;
1420         int check = 0;
1421
1422         spin_lock(&ci->i_ceph_lock);
1423         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1424              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1425         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1426                 /* nevermind! */
1427                 spin_unlock(&ci->i_ceph_lock);
1428                 goto out;
1429         }
1430         orig_gen = ci->i_rdcache_gen;
1431         spin_unlock(&ci->i_ceph_lock);
1432
1433         truncate_inode_pages(&inode->i_data, 0);
1434
1435         spin_lock(&ci->i_ceph_lock);
1436         if (orig_gen == ci->i_rdcache_gen &&
1437             orig_gen == ci->i_rdcache_revoking) {
1438                 dout("invalidate_pages %p gen %d successful\n", inode,
1439                      ci->i_rdcache_gen);
1440                 ci->i_rdcache_revoking--;
1441                 check = 1;
1442         } else {
1443                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1444                      inode, orig_gen, ci->i_rdcache_gen,
1445                      ci->i_rdcache_revoking);
1446         }
1447         spin_unlock(&ci->i_ceph_lock);
1448
1449         if (check)
1450                 ceph_check_caps(ci, 0, NULL);
1451 out:
1452         iput(inode);
1453 }
1454
1455
1456 /*
1457  * called by trunc_wq;
1458  *
1459  * We also truncate in a separate thread as well.
1460  */
1461 static void ceph_vmtruncate_work(struct work_struct *work)
1462 {
1463         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1464                                                   i_vmtruncate_work);
1465         struct inode *inode = &ci->vfs_inode;
1466
1467         dout("vmtruncate_work %p\n", inode);
1468         __ceph_do_pending_vmtruncate(inode, true);
1469         iput(inode);
1470 }
1471
1472 /*
1473  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1474  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1475  */
1476 void ceph_queue_vmtruncate(struct inode *inode)
1477 {
1478         struct ceph_inode_info *ci = ceph_inode(inode);
1479
1480         ihold(inode);
1481         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1482                        &ci->i_vmtruncate_work)) {
1483                 dout("ceph_queue_vmtruncate %p\n", inode);
1484         } else {
1485                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1486                      inode, ci->i_truncate_pending);
1487                 iput(inode);
1488         }
1489 }
1490
1491 /*
1492  * Make sure any pending truncation is applied before doing anything
1493  * that may depend on it.
1494  */
1495 void __ceph_do_pending_vmtruncate(struct inode *inode, bool needlock)
1496 {
1497         struct ceph_inode_info *ci = ceph_inode(inode);
1498         u64 to;
1499         int wrbuffer_refs, finish = 0;
1500
1501 retry:
1502         spin_lock(&ci->i_ceph_lock);
1503         if (ci->i_truncate_pending == 0) {
1504                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1505                 spin_unlock(&ci->i_ceph_lock);
1506                 return;
1507         }
1508
1509         /*
1510          * make sure any dirty snapped pages are flushed before we
1511          * possibly truncate them.. so write AND block!
1512          */
1513         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1514                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1515                      inode);
1516                 spin_unlock(&ci->i_ceph_lock);
1517                 filemap_write_and_wait_range(&inode->i_data, 0,
1518                                              inode->i_sb->s_maxbytes);
1519                 goto retry;
1520         }
1521
1522         to = ci->i_truncate_size;
1523         wrbuffer_refs = ci->i_wrbuffer_ref;
1524         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1525              ci->i_truncate_pending, to);
1526         spin_unlock(&ci->i_ceph_lock);
1527
1528         if (needlock)
1529                 mutex_lock(&inode->i_mutex);
1530         truncate_inode_pages(inode->i_mapping, to);
1531         if (needlock)
1532                 mutex_unlock(&inode->i_mutex);
1533
1534         spin_lock(&ci->i_ceph_lock);
1535         if (to == ci->i_truncate_size) {
1536                 ci->i_truncate_pending = 0;
1537                 finish = 1;
1538         }
1539         spin_unlock(&ci->i_ceph_lock);
1540         if (!finish)
1541                 goto retry;
1542
1543         if (wrbuffer_refs == 0)
1544                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1545
1546         wake_up_all(&ci->i_cap_wq);
1547 }
1548
1549
1550 /*
1551  * symlinks
1552  */
1553 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1554 {
1555         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1556         nd_set_link(nd, ci->i_symlink);
1557         return NULL;
1558 }
1559
1560 static const struct inode_operations ceph_symlink_iops = {
1561         .readlink = generic_readlink,
1562         .follow_link = ceph_sym_follow_link,
1563         .setattr = ceph_setattr,
1564         .getattr = ceph_getattr,
1565         .setxattr = ceph_setxattr,
1566         .getxattr = ceph_getxattr,
1567         .listxattr = ceph_listxattr,
1568         .removexattr = ceph_removexattr,
1569 };
1570
1571 /*
1572  * setattr
1573  */
1574 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1575 {
1576         struct inode *inode = dentry->d_inode;
1577         struct ceph_inode_info *ci = ceph_inode(inode);
1578         struct inode *parent_inode;
1579         const unsigned int ia_valid = attr->ia_valid;
1580         struct ceph_mds_request *req;
1581         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1582         int issued;
1583         int release = 0, dirtied = 0;
1584         int mask = 0;
1585         int err = 0;
1586         int inode_dirty_flags = 0;
1587
1588         if (ceph_snap(inode) != CEPH_NOSNAP)
1589                 return -EROFS;
1590
1591         __ceph_do_pending_vmtruncate(inode, false);
1592
1593         err = inode_change_ok(inode, attr);
1594         if (err != 0)
1595                 return err;
1596
1597         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1598                                        USE_AUTH_MDS);
1599         if (IS_ERR(req))
1600                 return PTR_ERR(req);
1601
1602         spin_lock(&ci->i_ceph_lock);
1603         issued = __ceph_caps_issued(ci, NULL);
1604         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1605
1606         if (ia_valid & ATTR_UID) {
1607                 dout("setattr %p uid %d -> %d\n", inode,
1608                      from_kuid(&init_user_ns, inode->i_uid),
1609                      from_kuid(&init_user_ns, attr->ia_uid));
1610                 if (issued & CEPH_CAP_AUTH_EXCL) {
1611                         inode->i_uid = attr->ia_uid;
1612                         dirtied |= CEPH_CAP_AUTH_EXCL;
1613                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1614                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1615                         req->r_args.setattr.uid = cpu_to_le32(
1616                                 from_kuid(&init_user_ns, attr->ia_uid));
1617                         mask |= CEPH_SETATTR_UID;
1618                         release |= CEPH_CAP_AUTH_SHARED;
1619                 }
1620         }
1621         if (ia_valid & ATTR_GID) {
1622                 dout("setattr %p gid %d -> %d\n", inode,
1623                      from_kgid(&init_user_ns, inode->i_gid),
1624                      from_kgid(&init_user_ns, attr->ia_gid));
1625                 if (issued & CEPH_CAP_AUTH_EXCL) {
1626                         inode->i_gid = attr->ia_gid;
1627                         dirtied |= CEPH_CAP_AUTH_EXCL;
1628                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1629                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1630                         req->r_args.setattr.gid = cpu_to_le32(
1631                                 from_kgid(&init_user_ns, attr->ia_gid));
1632                         mask |= CEPH_SETATTR_GID;
1633                         release |= CEPH_CAP_AUTH_SHARED;
1634                 }
1635         }
1636         if (ia_valid & ATTR_MODE) {
1637                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1638                      attr->ia_mode);
1639                 if (issued & CEPH_CAP_AUTH_EXCL) {
1640                         inode->i_mode = attr->ia_mode;
1641                         dirtied |= CEPH_CAP_AUTH_EXCL;
1642                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1643                            attr->ia_mode != inode->i_mode) {
1644                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1645                         mask |= CEPH_SETATTR_MODE;
1646                         release |= CEPH_CAP_AUTH_SHARED;
1647                 }
1648         }
1649
1650         if (ia_valid & ATTR_ATIME) {
1651                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1652                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1653                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1654                 if (issued & CEPH_CAP_FILE_EXCL) {
1655                         ci->i_time_warp_seq++;
1656                         inode->i_atime = attr->ia_atime;
1657                         dirtied |= CEPH_CAP_FILE_EXCL;
1658                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1659                            timespec_compare(&inode->i_atime,
1660                                             &attr->ia_atime) < 0) {
1661                         inode->i_atime = attr->ia_atime;
1662                         dirtied |= CEPH_CAP_FILE_WR;
1663                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1664                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1665                         ceph_encode_timespec(&req->r_args.setattr.atime,
1666                                              &attr->ia_atime);
1667                         mask |= CEPH_SETATTR_ATIME;
1668                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1669                                 CEPH_CAP_FILE_WR;
1670                 }
1671         }
1672         if (ia_valid & ATTR_MTIME) {
1673                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1674                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1675                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1676                 if (issued & CEPH_CAP_FILE_EXCL) {
1677                         ci->i_time_warp_seq++;
1678                         inode->i_mtime = attr->ia_mtime;
1679                         dirtied |= CEPH_CAP_FILE_EXCL;
1680                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1681                            timespec_compare(&inode->i_mtime,
1682                                             &attr->ia_mtime) < 0) {
1683                         inode->i_mtime = attr->ia_mtime;
1684                         dirtied |= CEPH_CAP_FILE_WR;
1685                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1686                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1687                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1688                                              &attr->ia_mtime);
1689                         mask |= CEPH_SETATTR_MTIME;
1690                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1691                                 CEPH_CAP_FILE_WR;
1692                 }
1693         }
1694         if (ia_valid & ATTR_SIZE) {
1695                 dout("setattr %p size %lld -> %lld\n", inode,
1696                      inode->i_size, attr->ia_size);
1697                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1698                         err = -EINVAL;
1699                         goto out;
1700                 }
1701                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1702                     attr->ia_size > inode->i_size) {
1703                         inode->i_size = attr->ia_size;
1704                         inode->i_blocks =
1705                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1706                         inode->i_ctime = attr->ia_ctime;
1707                         ci->i_reported_size = attr->ia_size;
1708                         dirtied |= CEPH_CAP_FILE_EXCL;
1709                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1710                            attr->ia_size != inode->i_size) {
1711                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1712                         req->r_args.setattr.old_size =
1713                                 cpu_to_le64(inode->i_size);
1714                         mask |= CEPH_SETATTR_SIZE;
1715                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1716                                 CEPH_CAP_FILE_WR;
1717                 }
1718         }
1719
1720         /* these do nothing */
1721         if (ia_valid & ATTR_CTIME) {
1722                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1723                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1724                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1725                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1726                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1727                      only ? "ctime only" : "ignored");
1728                 inode->i_ctime = attr->ia_ctime;
1729                 if (only) {
1730                         /*
1731                          * if kernel wants to dirty ctime but nothing else,
1732                          * we need to choose a cap to dirty under, or do
1733                          * a almost-no-op setattr
1734                          */
1735                         if (issued & CEPH_CAP_AUTH_EXCL)
1736                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1737                         else if (issued & CEPH_CAP_FILE_EXCL)
1738                                 dirtied |= CEPH_CAP_FILE_EXCL;
1739                         else if (issued & CEPH_CAP_XATTR_EXCL)
1740                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1741                         else
1742                                 mask |= CEPH_SETATTR_CTIME;
1743                 }
1744         }
1745         if (ia_valid & ATTR_FILE)
1746                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1747
1748         if (dirtied) {
1749                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1750                 inode->i_ctime = CURRENT_TIME;
1751         }
1752
1753         release &= issued;
1754         spin_unlock(&ci->i_ceph_lock);
1755
1756         if (inode_dirty_flags)
1757                 __mark_inode_dirty(inode, inode_dirty_flags);
1758
1759         if (mask) {
1760                 req->r_inode = inode;
1761                 ihold(inode);
1762                 req->r_inode_drop = release;
1763                 req->r_args.setattr.mask = cpu_to_le32(mask);
1764                 req->r_num_caps = 1;
1765                 parent_inode = ceph_get_dentry_parent_inode(dentry);
1766                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1767                 iput(parent_inode);
1768         }
1769         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1770              ceph_cap_string(dirtied), mask);
1771
1772         ceph_mdsc_put_request(req);
1773         __ceph_do_pending_vmtruncate(inode, false);
1774         return err;
1775 out:
1776         spin_unlock(&ci->i_ceph_lock);
1777         ceph_mdsc_put_request(req);
1778         return err;
1779 }
1780
1781 /*
1782  * Verify that we have a lease on the given mask.  If not,
1783  * do a getattr against an mds.
1784  */
1785 int ceph_do_getattr(struct inode *inode, int mask)
1786 {
1787         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1788         struct ceph_mds_client *mdsc = fsc->mdsc;
1789         struct ceph_mds_request *req;
1790         int err;
1791
1792         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1793                 dout("do_getattr inode %p SNAPDIR\n", inode);
1794                 return 0;
1795         }
1796
1797         dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1798         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1799                 return 0;
1800
1801         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1802         if (IS_ERR(req))
1803                 return PTR_ERR(req);
1804         req->r_inode = inode;
1805         ihold(inode);
1806         req->r_num_caps = 1;
1807         req->r_args.getattr.mask = cpu_to_le32(mask);
1808         err = ceph_mdsc_do_request(mdsc, NULL, req);
1809         ceph_mdsc_put_request(req);
1810         dout("do_getattr result=%d\n", err);
1811         return err;
1812 }
1813
1814
1815 /*
1816  * Check inode permissions.  We verify we have a valid value for
1817  * the AUTH cap, then call the generic handler.
1818  */
1819 int ceph_permission(struct inode *inode, int mask)
1820 {
1821         int err;
1822
1823         if (mask & MAY_NOT_BLOCK)
1824                 return -ECHILD;
1825
1826         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1827
1828         if (!err)
1829                 err = generic_permission(inode, mask);
1830         return err;
1831 }
1832
1833 /*
1834  * Get all attributes.  Hopefully somedata we'll have a statlite()
1835  * and can limit the fields we require to be accurate.
1836  */
1837 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1838                  struct kstat *stat)
1839 {
1840         struct inode *inode = dentry->d_inode;
1841         struct ceph_inode_info *ci = ceph_inode(inode);
1842         int err;
1843
1844         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1845         if (!err) {
1846                 generic_fillattr(inode, stat);
1847                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1848                 if (ceph_snap(inode) != CEPH_NOSNAP)
1849                         stat->dev = ceph_snap(inode);
1850                 else
1851                         stat->dev = 0;
1852                 if (S_ISDIR(inode->i_mode)) {
1853                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1854                                                 RBYTES))
1855                                 stat->size = ci->i_rbytes;
1856                         else
1857                                 stat->size = ci->i_files + ci->i_subdirs;
1858                         stat->blocks = 0;
1859                         stat->blksize = 65536;
1860                 }
1861         }
1862         return err;
1863 }