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