SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ceph/inode.c

   1 #include "ceph_debug.h"
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/smp_lock.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/namei.h>
  11 #include <linux/writeback.h>
  12 #include <linux/vmalloc.h>
  13
  14 #include "super.h"
  15 #include "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_inode_invalidate_pages(struct work_struct *work);
  32
  33 /*
  34  * find or create an inode, given the ceph ino number
  35  */
  36 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  37 {
  38         struct inode *inode;
  39         ino_t t = ceph_vino_to_ino(vino);
  40
  41         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  42         if (inode == NULL)
  43                 return ERR_PTR(-ENOMEM);
  44         if (inode->i_state & I_NEW) {
  45                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  46                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  47                 unlock_new_inode(inode);
  48         }
  49
  50         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  51              vino.snap, inode);
  52         return inode;
  53 }
  54
  55 /*
  56  * get/constuct snapdir inode for a given directory
  57  */
  58 struct inode *ceph_get_snapdir(struct inode *parent)
  59 {
  60         struct ceph_vino vino = {
  61                 .ino = ceph_ino(parent),
  62                 .snap = CEPH_SNAPDIR,
  63         };
  64         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  65
  66         BUG_ON(!S_ISDIR(parent->i_mode));
  67         if (IS_ERR(inode))
  68                 return ERR_PTR(PTR_ERR(inode));
  69         inode->i_mode = parent->i_mode;
  70         inode->i_uid = parent->i_uid;
  71         inode->i_gid = parent->i_gid;
  72         inode->i_op = &ceph_dir_iops;
  73         inode->i_fop = &ceph_dir_fops;
  74         ceph_inode(inode)->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  75         return inode;
  76 }
  77
  78 const struct inode_operations ceph_file_iops = {
  79         .permission = ceph_permission,
  80         .setattr = ceph_setattr,
  81         .getattr = ceph_getattr,
  82         .setxattr = ceph_setxattr,
  83         .getxattr = ceph_getxattr,
  84         .listxattr = ceph_listxattr,
  85         .removexattr = ceph_removexattr,
  86 };
  87
  88
  89 /*
  90  * We use a 'frag tree' to keep track of the MDS's directory fragments
  91  * for a given inode (usually there is just a single fragment).  We
  92  * need to know when a child frag is delegated to a new MDS, or when
  93  * it is flagged as replicated, so we can direct our requests
  94  * accordingly.
  95  */
  96
  97 /*
  98  * find/create a frag in the tree
  99  */
 100 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 101                                                     u32 f)
 102 {
 103         struct rb_node **p;
 104         struct rb_node *parent = NULL;
 105         struct ceph_inode_frag *frag;
 106         int c;
 107
 108         p = &ci->i_fragtree.rb_node;
 109         while (*p) {
 110                 parent = *p;
 111                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 112                 c = ceph_frag_compare(f, frag->frag);
 113                 if (c < 0)
 114                         p = &(*p)->rb_left;
 115                 else if (c > 0)
 116                         p = &(*p)->rb_right;
 117                 else
 118                         return frag;
 119         }
 120
 121         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 122         if (!frag) {
 123                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 124                        "frag %x\n", &ci->vfs_inode,
 125                        ceph_vinop(&ci->vfs_inode), f);
 126                 return ERR_PTR(-ENOMEM);
 127         }
 128         frag->frag = f;
 129         frag->split_by = 0;
 130         frag->mds = -1;
 131         frag->ndist = 0;
 132
 133         rb_link_node(&frag->node, parent, p);
 134         rb_insert_color(&frag->node, &ci->i_fragtree);
 135
 136         dout("get_or_create_frag added %llx.%llx frag %x\n",
 137              ceph_vinop(&ci->vfs_inode), f);
 138         return frag;
 139 }
 140
 141 /*
 142  * find a specific frag @f
 143  */
 144 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 145 {
 146         struct rb_node *n = ci->i_fragtree.rb_node;
 147
 148         while (n) {
 149                 struct ceph_inode_frag *frag =
 150                         rb_entry(n, struct ceph_inode_frag, node);
 151                 int c = ceph_frag_compare(f, frag->frag);
 152                 if (c < 0)
 153                         n = n->rb_left;
 154                 else if (c > 0)
 155                         n = n->rb_right;
 156                 else
 157                         return frag;
 158         }
 159         return NULL;
 160 }
 161
 162 /*
 163  * Choose frag containing the given value @v.  If @pfrag is
 164  * specified, copy the frag delegation info to the caller if
 165  * it is present.
 166  */
 167 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 168                      struct ceph_inode_frag *pfrag,
 169                      int *found)
 170 {
 171         u32 t = ceph_frag_make(0, 0);
 172         struct ceph_inode_frag *frag;
 173         unsigned nway, i;
 174         u32 n;
 175
 176         if (found)
 177                 *found = 0;
 178
 179         mutex_lock(&ci->i_fragtree_mutex);
 180         while (1) {
 181                 WARN_ON(!ceph_frag_contains_value(t, v));
 182                 frag = __ceph_find_frag(ci, t);
 183                 if (!frag)
 184                         break; /* t is a leaf */
 185                 if (frag->split_by == 0) {
 186                         if (pfrag)
 187                                 memcpy(pfrag, frag, sizeof(*pfrag));
 188                         if (found)
 189                                 *found = 1;
 190                         break;
 191                 }
 192
 193                 /* choose child */
 194                 nway = 1 << frag->split_by;
 195                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 196                      frag->split_by, nway);
 197                 for (i = 0; i < nway; i++) {
 198                         n = ceph_frag_make_child(t, frag->split_by, i);
 199                         if (ceph_frag_contains_value(n, v)) {
 200                                 t = n;
 201                                 break;
 202                         }
 203                 }
 204                 BUG_ON(i == nway);
 205         }
 206         dout("choose_frag(%x) = %x\n", v, t);
 207
 208         mutex_unlock(&ci->i_fragtree_mutex);
 209         return t;
 210 }
 211
 212 /*
 213  * Process dirfrag (delegation) info from the mds.  Include leaf
 214  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 215  * branches/splits are included in i_fragtree)
 216  */
 217 static int ceph_fill_dirfrag(struct inode *inode,
 218                              struct ceph_mds_reply_dirfrag *dirinfo)
 219 {
 220         struct ceph_inode_info *ci = ceph_inode(inode);
 221         struct ceph_inode_frag *frag;
 222         u32 id = le32_to_cpu(dirinfo->frag);
 223         int mds = le32_to_cpu(dirinfo->auth);
 224         int ndist = le32_to_cpu(dirinfo->ndist);
 225         int i;
 226         int err = 0;
 227
 228         mutex_lock(&ci->i_fragtree_mutex);
 229         if (ndist == 0) {
 230                 /* no delegation info needed. */
 231                 frag = __ceph_find_frag(ci, id);
 232                 if (!frag)
 233                         goto out;
 234                 if (frag->split_by == 0) {
 235                         /* tree leaf, remove */
 236                         dout("fill_dirfrag removed %llx.%llx frag %x"
 237                              " (no ref)\n", ceph_vinop(inode), id);
 238                         rb_erase(&frag->node, &ci->i_fragtree);
 239                         kfree(frag);
 240                 } else {
 241                         /* tree branch, keep and clear */
 242                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 243                              " referral\n", ceph_vinop(inode), id);
 244                         frag->mds = -1;
 245                         frag->ndist = 0;
 246                 }
 247                 goto out;
 248         }
 249
 250
 251         /* find/add this frag to store mds delegation info */
 252         frag = __get_or_create_frag(ci, id);
 253         if (IS_ERR(frag)) {
 254                 /* this is not the end of the world; we can continue
 255                    with bad/inaccurate delegation info */
 256                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 257                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 258                 err = -ENOMEM;
 259                 goto out;
 260         }
 261
 262         frag->mds = mds;
 263         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 264         for (i = 0; i < frag->ndist; i++)
 265                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 266         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 267              ceph_vinop(inode), frag->frag, frag->ndist);
 268
 269 out:
 270         mutex_unlock(&ci->i_fragtree_mutex);
 271         return err;
 272 }
 273
 274
 275 /*
 276  * initialize a newly allocated inode.
 277  */
 278 struct inode *ceph_alloc_inode(struct super_block *sb)
 279 {
 280         struct ceph_inode_info *ci;
 281         int i;
 282
 283         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 284         if (!ci)
 285                 return NULL;
 286
 287         dout("alloc_inode %p\n", &ci->vfs_inode);
 288
 289         ci->i_version = 0;
 290         ci->i_time_warp_seq = 0;
 291         ci->i_ceph_flags = 0;
 292         ci->i_release_count = 0;
 293         ci->i_symlink = NULL;
 294
 295         ci->i_fragtree = RB_ROOT;
 296         mutex_init(&ci->i_fragtree_mutex);
 297
 298         ci->i_xattrs.blob = NULL;
 299         ci->i_xattrs.prealloc_blob = NULL;
 300         ci->i_xattrs.dirty = false;
 301         ci->i_xattrs.index = RB_ROOT;
 302         ci->i_xattrs.count = 0;
 303         ci->i_xattrs.names_size = 0;
 304         ci->i_xattrs.vals_size = 0;
 305         ci->i_xattrs.version = 0;
 306         ci->i_xattrs.index_version = 0;
 307
 308         ci->i_caps = RB_ROOT;
 309         ci->i_auth_cap = NULL;
 310         ci->i_dirty_caps = 0;
 311         ci->i_flushing_caps = 0;
 312         INIT_LIST_HEAD(&ci->i_dirty_item);
 313         INIT_LIST_HEAD(&ci->i_flushing_item);
 314         ci->i_cap_flush_seq = 0;
 315         ci->i_cap_flush_last_tid = 0;
 316         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 317         init_waitqueue_head(&ci->i_cap_wq);
 318         ci->i_hold_caps_min = 0;
 319         ci->i_hold_caps_max = 0;
 320         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 321         ci->i_cap_exporting_mds = 0;
 322         ci->i_cap_exporting_mseq = 0;
 323         ci->i_cap_exporting_issued = 0;
 324         INIT_LIST_HEAD(&ci->i_cap_snaps);
 325         ci->i_head_snapc = NULL;
 326         ci->i_snap_caps = 0;
 327
 328         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 329                 ci->i_nr_by_mode[i] = 0;
 330
 331         ci->i_truncate_seq = 0;
 332         ci->i_truncate_size = 0;
 333         ci->i_truncate_pending = 0;
 334
 335         ci->i_max_size = 0;
 336         ci->i_reported_size = 0;
 337         ci->i_wanted_max_size = 0;
 338         ci->i_requested_max_size = 0;
 339
 340         ci->i_pin_ref = 0;
 341         ci->i_rd_ref = 0;
 342         ci->i_rdcache_ref = 0;
 343         ci->i_wr_ref = 0;
 344         ci->i_wrbuffer_ref = 0;
 345         ci->i_wrbuffer_ref_head = 0;
 346         ci->i_shared_gen = 0;
 347         ci->i_rdcache_gen = 0;
 348         ci->i_rdcache_revoking = 0;
 349
 350         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 351         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 352         spin_lock_init(&ci->i_unsafe_lock);
 353
 354         ci->i_snap_realm = NULL;
 355         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 356         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 357
 358         INIT_WORK(&ci->i_wb_work, ceph_inode_writeback);
 359         INIT_WORK(&ci->i_pg_inv_work, ceph_inode_invalidate_pages);
 360
 361         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 362
 363         return &ci->vfs_inode;
 364 }
 365
 366 void ceph_destroy_inode(struct inode *inode)
 367 {
 368         struct ceph_inode_info *ci = ceph_inode(inode);
 369         struct ceph_inode_frag *frag;
 370         struct rb_node *n;
 371
 372         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 373
 374         ceph_queue_caps_release(inode);
 375
 376         kfree(ci->i_symlink);
 377         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 378                 frag = rb_entry(n, struct ceph_inode_frag, node);
 379                 rb_erase(n, &ci->i_fragtree);
 380                 kfree(frag);
 381         }
 382
 383         __ceph_destroy_xattrs(ci);
 384         ceph_buffer_put(ci->i_xattrs.blob);
 385         ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 386
 387         kmem_cache_free(ceph_inode_cachep, ci);
 388 }
 389
 390
 391 /*
 392  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 393  * careful because either the client or MDS may have more up to date
 394  * info, depending on which capabilities are held, and whether
 395  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 396  * and size are monotonically increasing, except when utimes() or
 397  * truncate() increments the corresponding _seq values.)
 398  */
 399 int ceph_fill_file_size(struct inode *inode, int issued,
 400                         u32 truncate_seq, u64 truncate_size, u64 size)
 401 {
 402         struct ceph_inode_info *ci = ceph_inode(inode);
 403         int queue_trunc = 0;
 404
 405         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 406             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 407                 dout("size %lld -> %llu\n", inode->i_size, size);
 408                 inode->i_size = size;
 409                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 410                 ci->i_reported_size = size;
 411                 if (truncate_seq != ci->i_truncate_seq) {
 412                         dout("truncate_seq %u -> %u\n",
 413                              ci->i_truncate_seq, truncate_seq);
 414                         ci->i_truncate_seq = truncate_seq;
 415                         if (issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
 416                                       CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
 417                                       CEPH_CAP_FILE_EXCL)) {
 418                                 ci->i_truncate_pending++;
 419                                 queue_trunc = 1;
 420                         }
 421                 }
 422         }
 423         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 424             ci->i_truncate_size != truncate_size) {
 425                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 426                      truncate_size);
 427                 ci->i_truncate_size = truncate_size;
 428         }
 429         return queue_trunc;
 430 }
 431
 432 void ceph_fill_file_time(struct inode *inode, int issued,
 433                          u64 time_warp_seq, struct timespec *ctime,
 434                          struct timespec *mtime, struct timespec *atime)
 435 {
 436         struct ceph_inode_info *ci = ceph_inode(inode);
 437         int warn = 0;
 438
 439         if (issued & (CEPH_CAP_FILE_EXCL|
 440                       CEPH_CAP_FILE_WR|
 441                       CEPH_CAP_FILE_BUFFER)) {
 442                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 443                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 444                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 445                              ctime->tv_sec, ctime->tv_nsec);
 446                         inode->i_ctime = *ctime;
 447                 }
 448                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 449                         /* the MDS did a utimes() */
 450                         dout("mtime %ld.%09ld -> %ld.%09ld "
 451                              "tw %d -> %d\n",
 452                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 453                              mtime->tv_sec, mtime->tv_nsec,
 454                              ci->i_time_warp_seq, (int)time_warp_seq);
 455
 456                         inode->i_mtime = *mtime;
 457                         inode->i_atime = *atime;
 458                         ci->i_time_warp_seq = time_warp_seq;
 459                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 460                         /* nobody did utimes(); take the max */
 461                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 462                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 463                                      inode->i_mtime.tv_sec,
 464                                      inode->i_mtime.tv_nsec,
 465                                      mtime->tv_sec, mtime->tv_nsec);
 466                                 inode->i_mtime = *mtime;
 467                         }
 468                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 469                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 470                                      inode->i_atime.tv_sec,
 471                                      inode->i_atime.tv_nsec,
 472                                      atime->tv_sec, atime->tv_nsec);
 473                                 inode->i_atime = *atime;
 474                         }
 475                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 476                         /* we did a utimes(); ignore mds values */
 477                 } else {
 478                         warn = 1;
 479                 }
 480         } else {
 481                 /* we have no write caps; whatever the MDS says is true */
 482                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 483                         inode->i_ctime = *ctime;
 484                         inode->i_mtime = *mtime;
 485                         inode->i_atime = *atime;
 486                         ci->i_time_warp_seq = time_warp_seq;
 487                 } else {
 488                         warn = 1;
 489                 }
 490         }
 491         if (warn) /* time_warp_seq shouldn't go backwards */
 492                 dout("%p mds time_warp_seq %llu < %u\n",
 493                      inode, time_warp_seq, ci->i_time_warp_seq);
 494 }
 495
 496 /*
 497  * Populate an inode based on info from mds.  May be called on new or
 498  * existing inodes.
 499  */
 500 static int fill_inode(struct inode *inode,
 501                       struct ceph_mds_reply_info_in *iinfo,
 502                       struct ceph_mds_reply_dirfrag *dirinfo,
 503                       struct ceph_mds_session *session,
 504                       unsigned long ttl_from, int cap_fmode,
 505                       struct ceph_cap_reservation *caps_reservation)
 506 {
 507         struct ceph_mds_reply_inode *info = iinfo->in;
 508         struct ceph_inode_info *ci = ceph_inode(inode);
 509         int i;
 510         int issued, implemented;
 511         struct timespec mtime, atime, ctime;
 512         u32 nsplits;
 513         struct ceph_buffer *xattr_blob = NULL;
 514         int err = 0;
 515         int queue_trunc = 0;
 516
 517         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 518              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 519              ci->i_version);
 520
 521         /*
 522          * prealloc xattr data, if it looks like we'll need it.  only
 523          * if len > 4 (meaning there are actually xattrs; the first 4
 524          * bytes are the xattr count).
 525          */
 526         if (iinfo->xattr_len > 4) {
 527                 xattr_blob = ceph_buffer_new_alloc(iinfo->xattr_len, GFP_NOFS);
 528                 if (!xattr_blob)
 529                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 530                                iinfo->xattr_len);
 531         }
 532
 533         spin_lock(&inode->i_lock);
 534
 535         /*
 536          * provided version will be odd if inode value is projected,
 537          * even if stable.  skip the update if we have a newer info
 538          * (e.g., due to inode info racing form multiple MDSs), or if
 539          * we are getting projected (unstable) inode info.
 540          */
 541         if (le64_to_cpu(info->version) > 0 &&
 542             (ci->i_version & ~1) > le64_to_cpu(info->version))
 543                 goto no_change;
 544
 545         issued = __ceph_caps_issued(ci, &implemented);
 546         issued |= implemented | __ceph_caps_dirty(ci);
 547
 548         /* update inode */
 549         ci->i_version = le64_to_cpu(info->version);
 550         inode->i_version++;
 551         inode->i_rdev = le32_to_cpu(info->rdev);
 552
 553         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
 554                 inode->i_mode = le32_to_cpu(info->mode);
 555                 inode->i_uid = le32_to_cpu(info->uid);
 556                 inode->i_gid = le32_to_cpu(info->gid);
 557                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 558                      inode->i_uid, inode->i_gid);
 559         }
 560
 561         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
 562                 inode->i_nlink = le32_to_cpu(info->nlink);
 563
 564         /* be careful with mtime, atime, size */
 565         ceph_decode_timespec(&atime, &info->atime);
 566         ceph_decode_timespec(&mtime, &info->mtime);
 567         ceph_decode_timespec(&ctime, &info->ctime);
 568         queue_trunc = ceph_fill_file_size(inode, issued,
 569                                           le32_to_cpu(info->truncate_seq),
 570                                           le64_to_cpu(info->truncate_size),
 571                                           S_ISDIR(inode->i_mode) ?
 572                                           ci->i_rbytes :
 573                                           le64_to_cpu(info->size));
 574         ceph_fill_file_time(inode, issued,
 575                             le32_to_cpu(info->time_warp_seq),
 576                             &ctime, &mtime, &atime);
 577
 578         ci->i_max_size = le64_to_cpu(info->max_size);
 579         ci->i_layout = info->layout;
 580         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 581
 582         /* xattrs */
 583         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 584         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
 585             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 586                 if (ci->i_xattrs.blob)
 587                         ceph_buffer_put(ci->i_xattrs.blob);
 588                 ci->i_xattrs.blob = xattr_blob;
 589                 if (xattr_blob)
 590                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 591                                iinfo->xattr_data, iinfo->xattr_len);
 592                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 593         }
 594
 595         inode->i_mapping->a_ops = &ceph_aops;
 596         inode->i_mapping->backing_dev_info =
 597                 &ceph_client(inode->i_sb)->backing_dev_info;
 598
 599         switch (inode->i_mode & S_IFMT) {
 600         case S_IFIFO:
 601         case S_IFBLK:
 602         case S_IFCHR:
 603         case S_IFSOCK:
 604                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 605                 inode->i_op = &ceph_file_iops;
 606                 break;
 607         case S_IFREG:
 608                 inode->i_op = &ceph_file_iops;
 609                 inode->i_fop = &ceph_file_fops;
 610                 break;
 611         case S_IFLNK:
 612                 inode->i_op = &ceph_symlink_iops;
 613                 if (!ci->i_symlink) {
 614                         int symlen = iinfo->symlink_len;
 615                         char *sym;
 616
 617                         BUG_ON(symlen != inode->i_size);
 618                         spin_unlock(&inode->i_lock);
 619
 620                         err = -ENOMEM;
 621                         sym = kmalloc(symlen+1, GFP_NOFS);
 622                         if (!sym)
 623                                 goto out;
 624                         memcpy(sym, iinfo->symlink, symlen);
 625                         sym[symlen] = 0;
 626
 627                         spin_lock(&inode->i_lock);
 628                         if (!ci->i_symlink)
 629                                 ci->i_symlink = sym;
 630                         else
 631                                 kfree(sym); /* lost a race */
 632                 }
 633                 break;
 634         case S_IFDIR:
 635                 inode->i_op = &ceph_dir_iops;
 636                 inode->i_fop = &ceph_dir_fops;
 637
 638                 ci->i_files = le64_to_cpu(info->files);
 639                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 640                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 641                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 642                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 643                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 644
 645                 /* set dir completion flag? */
 646                 if (ci->i_files == 0 && ci->i_subdirs == 0 &&
 647                     ceph_snap(inode) == CEPH_NOSNAP &&
 648                     (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED)) {
 649                         dout(" marking %p complete (empty)\n", inode);
 650                         ci->i_ceph_flags |= CEPH_I_COMPLETE;
 651                         ci->i_max_offset = 2;
 652                 }
 653
 654                 /* it may be better to set st_size in getattr instead? */
 655                 if (ceph_test_opt(ceph_client(inode->i_sb), RBYTES))
 656                         inode->i_size = ci->i_rbytes;
 657                 break;
 658         default:
 659                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 660                        ceph_vinop(inode), inode->i_mode);
 661         }
 662
 663 no_change:
 664         spin_unlock(&inode->i_lock);
 665
 666         /* queue truncate if we saw i_size decrease */
 667         if (queue_trunc)
 668                 if (queue_work(ceph_client(inode->i_sb)->trunc_wq,
 669                                &ci->i_vmtruncate_work))
 670                         igrab(inode);
 671
 672         /* populate frag tree */
 673         /* FIXME: move me up, if/when version reflects fragtree changes */
 674         nsplits = le32_to_cpu(info->fragtree.nsplits);
 675         mutex_lock(&ci->i_fragtree_mutex);
 676         for (i = 0; i < nsplits; i++) {
 677                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
 678                 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
 679
 680                 if (IS_ERR(frag))
 681                         continue;
 682                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
 683                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 684         }
 685         mutex_unlock(&ci->i_fragtree_mutex);
 686
 687         /* were we issued a capability? */
 688         if (info->cap.caps) {
 689                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 690                         ceph_add_cap(inode, session,
 691                                      le64_to_cpu(info->cap.cap_id),
 692                                      cap_fmode,
 693                                      le32_to_cpu(info->cap.caps),
 694                                      le32_to_cpu(info->cap.wanted),
 695                                      le32_to_cpu(info->cap.seq),
 696                                      le32_to_cpu(info->cap.mseq),
 697                                      le64_to_cpu(info->cap.realm),
 698                                      info->cap.flags,
 699                                      caps_reservation);
 700                 } else {
 701                         spin_lock(&inode->i_lock);
 702                         dout(" %p got snap_caps %s\n", inode,
 703                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 704                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 705                         if (cap_fmode >= 0)
 706                                 __ceph_get_fmode(ci, cap_fmode);
 707                         spin_unlock(&inode->i_lock);
 708                 }
 709         }
 710
 711         /* update delegation info? */
 712         if (dirinfo)
 713                 ceph_fill_dirfrag(inode, dirinfo);
 714
 715         err = 0;
 716
 717 out:
 718         ceph_buffer_put(xattr_blob);
 719         return err;
 720 }
 721
 722 /*
 723  * caller should hold session s_mutex.
 724  */
 725 static void update_dentry_lease(struct dentry *dentry,
 726                                 struct ceph_mds_reply_lease *lease,
 727                                 struct ceph_mds_session *session,
 728                                 unsigned long from_time)
 729 {
 730         struct ceph_dentry_info *di = ceph_dentry(dentry);
 731         long unsigned duration = le32_to_cpu(lease->duration_ms);
 732         long unsigned ttl = from_time + (duration * HZ) / 1000;
 733         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 734         struct inode *dir;
 735
 736         /* only track leases on regular dentries */
 737         if (dentry->d_op != &ceph_dentry_ops)
 738                 return;
 739
 740         spin_lock(&dentry->d_lock);
 741         dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
 742              dentry, le16_to_cpu(lease->mask), duration, ttl);
 743
 744         /* make lease_rdcache_gen match directory */
 745         dir = dentry->d_parent->d_inode;
 746         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 747
 748         if (lease->mask == 0)
 749                 goto out_unlock;
 750
 751         if (di->lease_gen == session->s_cap_gen &&
 752             time_before(ttl, dentry->d_time))
 753                 goto out_unlock;  /* we already have a newer lease. */
 754
 755         if (di->lease_session && di->lease_session != session)
 756                 goto out_unlock;
 757
 758         ceph_dentry_lru_touch(dentry);
 759
 760         if (!di->lease_session)
 761                 di->lease_session = ceph_get_mds_session(session);
 762         di->lease_gen = session->s_cap_gen;
 763         di->lease_seq = le32_to_cpu(lease->seq);
 764         di->lease_renew_after = half_ttl;
 765         di->lease_renew_from = 0;
 766         dentry->d_time = ttl;
 767 out_unlock:
 768         spin_unlock(&dentry->d_lock);
 769         return;
 770 }
 771
 772 /*
 773  * splice a dentry to an inode.
 774  * caller must hold directory i_mutex for this to be safe.
 775  *
 776  * we will only rehash the resulting dentry if @prehash is
 777  * true; @prehash will be set to false (for the benefit of
 778  * the caller) if we fail.
 779  */
 780 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 781                                     bool *prehash)
 782 {
 783         struct dentry *realdn;
 784
 785         /* dn must be unhashed */
 786         if (!d_unhashed(dn))
 787                 d_drop(dn);
 788         realdn = d_materialise_unique(dn, in);
 789         if (IS_ERR(realdn)) {
 790                 pr_err("splice_dentry error %p inode %p ino %llx.%llx\n",
 791                        dn, in, ceph_vinop(in));
 792                 if (prehash)
 793                         *prehash = false; /* don't rehash on error */
 794                 dn = realdn; /* note realdn contains the error */
 795                 goto out;
 796         } else if (realdn) {
 797                 dout("dn %p (%d) spliced with %p (%d) "
 798                      "inode %p ino %llx.%llx\n",
 799                      dn, atomic_read(&dn->d_count),
 800                      realdn, atomic_read(&realdn->d_count),
 801                      realdn->d_inode, ceph_vinop(realdn->d_inode));
 802                 dput(dn);
 803                 dn = realdn;
 804         } else {
 805                 BUG_ON(!ceph_dentry(dn));
 806
 807                 dout("dn %p attached to %p ino %llx.%llx\n",
 808                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
 809         }
 810         if ((!prehash || *prehash) && d_unhashed(dn))
 811                 d_rehash(dn);
 812 out:
 813         return dn;
 814 }
 815
 816 /*
 817  * Incorporate results into the local cache.  This is either just
 818  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
 819  * after a lookup).
 820  *
 821  * A reply may contain
 822  *         a directory inode along with a dentry.
 823  *  and/or a target inode
 824  *
 825  * Called with snap_rwsem (read).
 826  */
 827 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
 828                     struct ceph_mds_session *session)
 829 {
 830         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
 831         struct inode *in = NULL;
 832         struct ceph_mds_reply_inode *ininfo;
 833         struct ceph_vino vino;
 834         int i = 0;
 835         int err = 0;
 836
 837         dout("fill_trace %p is_dentry %d is_target %d\n", req,
 838              rinfo->head->is_dentry, rinfo->head->is_target);
 839
 840 #if 0
 841         /*
 842          * Debugging hook:
 843          *
 844          * If we resend completed ops to a recovering mds, we get no
 845          * trace.  Since that is very rare, pretend this is the case
 846          * to ensure the 'no trace' handlers in the callers behave.
 847          *
 848          * Fill in inodes unconditionally to avoid breaking cap
 849          * invariants.
 850          */
 851         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
 852                 pr_info("fill_trace faking empty trace on %lld %s\n",
 853                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
 854                 if (rinfo->head->is_dentry) {
 855                         rinfo->head->is_dentry = 0;
 856                         err = fill_inode(req->r_locked_dir,
 857                                          &rinfo->diri, rinfo->dirfrag,
 858                                          session, req->r_request_started, -1);
 859                 }
 860                 if (rinfo->head->is_target) {
 861                         rinfo->head->is_target = 0;
 862                         ininfo = rinfo->targeti.in;
 863                         vino.ino = le64_to_cpu(ininfo->ino);
 864                         vino.snap = le64_to_cpu(ininfo->snapid);
 865                         in = ceph_get_inode(sb, vino);
 866                         err = fill_inode(in, &rinfo->targeti, NULL,
 867                                          session, req->r_request_started,
 868                                          req->r_fmode);
 869                         iput(in);
 870                 }
 871         }
 872 #endif
 873
 874         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
 875                 dout("fill_trace reply is empty!\n");
 876                 if (rinfo->head->result == 0 && req->r_locked_dir) {
 877                         struct ceph_inode_info *ci =
 878                                 ceph_inode(req->r_locked_dir);
 879                         dout(" clearing %p complete (empty trace)\n",
 880                              req->r_locked_dir);
 881                         ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
 882                         ci->i_release_count++;
 883                 }
 884                 return 0;
 885         }
 886
 887         if (rinfo->head->is_dentry) {
 888                 /*
 889                  * lookup link rename   : null -> possibly existing inode
 890                  * mknod symlink mkdir  : null -> new inode
 891                  * unlink               : linked -> null
 892                  */
 893                 struct inode *dir = req->r_locked_dir;
 894                 struct dentry *dn = req->r_dentry;
 895                 bool have_dir_cap, have_lease;
 896
 897                 BUG_ON(!dn);
 898                 BUG_ON(!dir);
 899                 BUG_ON(dn->d_parent->d_inode != dir);
 900                 BUG_ON(ceph_ino(dir) !=
 901                        le64_to_cpu(rinfo->diri.in->ino));
 902                 BUG_ON(ceph_snap(dir) !=
 903                        le64_to_cpu(rinfo->diri.in->snapid));
 904
 905                 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
 906                                  session, req->r_request_started, -1,
 907                                  &req->r_caps_reservation);
 908                 if (err < 0)
 909                         return err;
 910
 911                 /* do we have a lease on the whole dir? */
 912                 have_dir_cap =
 913                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
 914                          CEPH_CAP_FILE_SHARED);
 915
 916                 /* do we have a dn lease? */
 917                 have_lease = have_dir_cap ||
 918                         (le16_to_cpu(rinfo->dlease->mask) &
 919                          CEPH_LOCK_DN);
 920
 921                 if (!have_lease)
 922                         dout("fill_trace  no dentry lease or dir cap\n");
 923
 924                 /* rename? */
 925                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
 926                         dout(" src %p '%.*s' dst %p '%.*s'\n",
 927                              req->r_old_dentry,
 928                              req->r_old_dentry->d_name.len,
 929                              req->r_old_dentry->d_name.name,
 930                              dn, dn->d_name.len, dn->d_name.name);
 931                         dout("fill_trace doing d_move %p -> %p\n",
 932                              req->r_old_dentry, dn);
 933                         d_move(req->r_old_dentry, dn);
 934                         dout(" src %p '%.*s' dst %p '%.*s'\n",
 935                              req->r_old_dentry,
 936                              req->r_old_dentry->d_name.len,
 937                              req->r_old_dentry->d_name.name,
 938                              dn, dn->d_name.len, dn->d_name.name);
 939                         /* take overwritten dentry's readdir offset */
 940                         ceph_dentry(req->r_old_dentry)->offset =
 941                                 ceph_dentry(dn)->offset;
 942                         dn = req->r_old_dentry;  /* use old_dentry */
 943                         in = dn->d_inode;
 944                 }
 945
 946                 /* null dentry? */
 947                 if (!rinfo->head->is_target) {
 948                         dout("fill_trace null dentry\n");
 949                         if (dn->d_inode) {
 950                                 dout("d_delete %p\n", dn);
 951                                 d_delete(dn);
 952                         } else {
 953                                 dout("d_instantiate %p NULL\n", dn);
 954                                 d_instantiate(dn, NULL);
 955                                 if (have_lease && d_unhashed(dn))
 956                                         d_rehash(dn);
 957                                 update_dentry_lease(dn, rinfo->dlease,
 958                                                     session,
 959                                                     req->r_request_started);
 960                         }
 961                         goto done;
 962                 }
 963
 964                 /* attach proper inode */
 965                 ininfo = rinfo->targeti.in;
 966                 vino.ino = le64_to_cpu(ininfo->ino);
 967                 vino.snap = le64_to_cpu(ininfo->snapid);
 968                 if (!dn->d_inode) {
 969                         in = ceph_get_inode(sb, vino);
 970                         if (IS_ERR(in)) {
 971                                 pr_err("fill_trace bad get_inode "
 972                                        "%llx.%llx\n", vino.ino, vino.snap);
 973                                 err = PTR_ERR(in);
 974                                 d_delete(dn);
 975                                 goto done;
 976                         }
 977                         dn = splice_dentry(dn, in, &have_lease);
 978                         if (IS_ERR(dn)) {
 979                                 err = PTR_ERR(dn);
 980                                 goto done;
 981                         }
 982                         req->r_dentry = dn;  /* may have spliced */
 983                         igrab(in);
 984                 } else if (ceph_ino(in) == vino.ino &&
 985                            ceph_snap(in) == vino.snap) {
 986                         igrab(in);
 987                 } else {
 988                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
 989                              dn, in, ceph_ino(in), ceph_snap(in),
 990                              vino.ino, vino.snap);
 991                         have_lease = false;
 992                         in = NULL;
 993                 }
 994
 995                 if (have_lease)
 996                         update_dentry_lease(dn, rinfo->dlease, session,
 997                                             req->r_request_started);
 998                 dout(" final dn %p\n", dn);
 999                 i++;
1000         } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1001                    req->r_op == CEPH_MDS_OP_MKSNAP) {
1002                 struct dentry *dn = req->r_dentry;
1003
1004                 /* fill out a snapdir LOOKUPSNAP dentry */
1005                 BUG_ON(!dn);
1006                 BUG_ON(!req->r_locked_dir);
1007                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
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                 if (IS_ERR(in)) {
1013                         pr_err("fill_inode get_inode badness %llx.%llx\n",
1014                                vino.ino, vino.snap);
1015                         err = PTR_ERR(in);
1016                         d_delete(dn);
1017                         goto done;
1018                 }
1019                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1020                 dn = splice_dentry(dn, in, NULL);
1021                 if (IS_ERR(dn)) {
1022                         err = PTR_ERR(dn);
1023                         goto done;
1024                 }
1025                 req->r_dentry = dn;  /* may have spliced */
1026                 igrab(in);
1027                 rinfo->head->is_dentry = 1;  /* fool notrace handlers */
1028         }
1029
1030         if (rinfo->head->is_target) {
1031                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1032                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1033
1034                 if (in == NULL || ceph_ino(in) != vino.ino ||
1035                     ceph_snap(in) != vino.snap) {
1036                         in = ceph_get_inode(sb, vino);
1037                         if (IS_ERR(in)) {
1038                                 err = PTR_ERR(in);
1039                                 goto done;
1040                         }
1041                 }
1042                 req->r_target_inode = in;
1043
1044                 err = fill_inode(in,
1045                                  &rinfo->targeti, NULL,
1046                                  session, req->r_request_started,
1047                                  (le32_to_cpu(rinfo->head->result) == 0) ?
1048                                  req->r_fmode : -1,
1049                                  &req->r_caps_reservation);
1050                 if (err < 0) {
1051                         pr_err("fill_inode badness %p %llx.%llx\n",
1052                                in, ceph_vinop(in));
1053                         goto done;
1054                 }
1055         }
1056
1057 done:
1058         dout("fill_trace done err=%d\n", err);
1059         return err;
1060 }
1061
1062 /*
1063  * Prepopulate our cache with readdir results, leases, etc.
1064  */
1065 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1066                              struct ceph_mds_session *session)
1067 {
1068         struct dentry *parent = req->r_dentry;
1069         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1070         struct qstr dname;
1071         struct dentry *dn;
1072         struct inode *in;
1073         int err = 0, i;
1074         struct inode *snapdir = NULL;
1075         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1076         u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1077         struct ceph_dentry_info *di;
1078
1079         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1080                 snapdir = ceph_get_snapdir(parent->d_inode);
1081                 parent = d_find_alias(snapdir);
1082                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1083                      rinfo->dir_nr, parent);
1084         } else {
1085                 dout("readdir_prepopulate %d items under dn %p\n",
1086                      rinfo->dir_nr, parent);
1087                 if (rinfo->dir_dir)
1088                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1089         }
1090
1091         for (i = 0; i < rinfo->dir_nr; i++) {
1092                 struct ceph_vino vino;
1093
1094                 dname.name = rinfo->dir_dname[i];
1095                 dname.len = rinfo->dir_dname_len[i];
1096                 dname.hash = full_name_hash(dname.name, dname.len);
1097
1098                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1099                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1100
1101 retry_lookup:
1102                 dn = d_lookup(parent, &dname);
1103                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1104                      parent, dname.len, dname.name, dn);
1105
1106                 if (!dn) {
1107                         dn = d_alloc(parent, &dname);
1108                         dout("d_alloc %p '%.*s' = %p\n", parent,
1109                              dname.len, dname.name, dn);
1110                         if (dn == NULL) {
1111                                 dout("d_alloc badness\n");
1112                                 err = -ENOMEM;
1113                                 goto out;
1114                         }
1115                         err = ceph_init_dentry(dn);
1116                         if (err < 0)
1117                                 goto out;
1118                 } else if (dn->d_inode &&
1119                            (ceph_ino(dn->d_inode) != vino.ino ||
1120                             ceph_snap(dn->d_inode) != vino.snap)) {
1121                         dout(" dn %p points to wrong inode %p\n",
1122                              dn, dn->d_inode);
1123                         d_delete(dn);
1124                         dput(dn);
1125                         goto retry_lookup;
1126                 } else {
1127                         /* reorder parent's d_subdirs */
1128                         spin_lock(&dcache_lock);
1129                         spin_lock(&dn->d_lock);
1130                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1131                         spin_unlock(&dn->d_lock);
1132                         spin_unlock(&dcache_lock);
1133                 }
1134
1135                 di = dn->d_fsdata;
1136                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1137
1138                 /* inode */
1139                 if (dn->d_inode) {
1140                         in = dn->d_inode;
1141                 } else {
1142                         in = ceph_get_inode(parent->d_sb, vino);
1143                         if (in == NULL) {
1144                                 dout("new_inode badness\n");
1145                                 d_delete(dn);
1146                                 dput(dn);
1147                                 err = -ENOMEM;
1148                                 goto out;
1149                         }
1150                         dn = splice_dentry(dn, in, NULL);
1151                 }
1152
1153                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1154                                req->r_request_started, -1,
1155                                &req->r_caps_reservation) < 0) {
1156                         pr_err("fill_inode badness on %p\n", in);
1157                         dput(dn);
1158                         continue;
1159                 }
1160                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1161                                     req->r_session, req->r_request_started);
1162                 dput(dn);
1163         }
1164         req->r_did_prepopulate = true;
1165
1166 out:
1167         if (snapdir) {
1168                 iput(snapdir);
1169                 dput(parent);
1170         }
1171         dout("readdir_prepopulate done\n");
1172         return err;
1173 }
1174
1175 int ceph_inode_set_size(struct inode *inode, loff_t size)
1176 {
1177         struct ceph_inode_info *ci = ceph_inode(inode);
1178         int ret = 0;
1179
1180         spin_lock(&inode->i_lock);
1181         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1182         inode->i_size = size;
1183         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1184
1185         /* tell the MDS if we are approaching max_size */
1186         if ((size << 1) >= ci->i_max_size &&
1187             (ci->i_reported_size << 1) < ci->i_max_size)
1188                 ret = 1;
1189
1190         spin_unlock(&inode->i_lock);
1191         return ret;
1192 }
1193
1194 /*
1195  * Write back inode data in a worker thread.  (This can't be done
1196  * in the message handler context.)
1197  */
1198 void ceph_inode_writeback(struct work_struct *work)
1199 {
1200         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1201                                                   i_wb_work);
1202         struct inode *inode = &ci->vfs_inode;
1203
1204         dout("writeback %p\n", inode);
1205         filemap_fdatawrite(&inode->i_data);
1206         iput(inode);
1207 }
1208
1209 /*
1210  * Invalidate inode pages in a worker thread.  (This can't be done
1211  * in the message handler context.)
1212  */
1213 static void ceph_inode_invalidate_pages(struct work_struct *work)
1214 {
1215         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1216                                                   i_pg_inv_work);
1217         struct inode *inode = &ci->vfs_inode;
1218         u32 orig_gen;
1219         int check = 0;
1220
1221         spin_lock(&inode->i_lock);
1222         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1223              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1224         if (ci->i_rdcache_gen == 0 ||
1225             ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1226                 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen);
1227                 /* nevermind! */
1228                 ci->i_rdcache_revoking = 0;
1229                 spin_unlock(&inode->i_lock);
1230                 goto out;
1231         }
1232         orig_gen = ci->i_rdcache_gen;
1233         spin_unlock(&inode->i_lock);
1234
1235         truncate_inode_pages(&inode->i_data, 0);
1236
1237         spin_lock(&inode->i_lock);
1238         if (orig_gen == ci->i_rdcache_gen) {
1239                 dout("invalidate_pages %p gen %d successful\n", inode,
1240                      ci->i_rdcache_gen);
1241                 ci->i_rdcache_gen = 0;
1242                 ci->i_rdcache_revoking = 0;
1243                 check = 1;
1244         } else {
1245                 dout("invalidate_pages %p gen %d raced, gen now %d\n",
1246                      inode, orig_gen, ci->i_rdcache_gen);
1247         }
1248         spin_unlock(&inode->i_lock);
1249
1250         if (check)
1251                 ceph_check_caps(ci, 0, NULL);
1252 out:
1253         iput(inode);
1254 }
1255
1256
1257 /*
1258  * called by trunc_wq; take i_mutex ourselves
1259  *
1260  * We also truncate in a separate thread as well.
1261  */
1262 void ceph_vmtruncate_work(struct work_struct *work)
1263 {
1264         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1265                                                   i_vmtruncate_work);
1266         struct inode *inode = &ci->vfs_inode;
1267
1268         dout("vmtruncate_work %p\n", inode);
1269         mutex_lock(&inode->i_mutex);
1270         __ceph_do_pending_vmtruncate(inode);
1271         mutex_unlock(&inode->i_mutex);
1272         iput(inode);
1273 }
1274
1275 /*
1276  * called with i_mutex held.
1277  *
1278  * Make sure any pending truncation is applied before doing anything
1279  * that may depend on it.
1280  */
1281 void __ceph_do_pending_vmtruncate(struct inode *inode)
1282 {
1283         struct ceph_inode_info *ci = ceph_inode(inode);
1284         u64 to;
1285         int wrbuffer_refs, wake = 0;
1286
1287 retry:
1288         spin_lock(&inode->i_lock);
1289         if (ci->i_truncate_pending == 0) {
1290                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1291                 spin_unlock(&inode->i_lock);
1292                 return;
1293         }
1294
1295         /*
1296          * make sure any dirty snapped pages are flushed before we
1297          * possibly truncate them.. so write AND block!
1298          */
1299         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1300                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1301                      inode);
1302                 spin_unlock(&inode->i_lock);
1303                 filemap_write_and_wait_range(&inode->i_data, 0,
1304                                              inode->i_sb->s_maxbytes);
1305                 goto retry;
1306         }
1307
1308         to = ci->i_truncate_size;
1309         wrbuffer_refs = ci->i_wrbuffer_ref;
1310         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1311              ci->i_truncate_pending, to);
1312         spin_unlock(&inode->i_lock);
1313
1314         truncate_inode_pages(inode->i_mapping, to);
1315
1316         spin_lock(&inode->i_lock);
1317         ci->i_truncate_pending--;
1318         if (ci->i_truncate_pending == 0)
1319                 wake = 1;
1320         spin_unlock(&inode->i_lock);
1321
1322         if (wrbuffer_refs == 0)
1323                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1324         if (wake)
1325                 wake_up(&ci->i_cap_wq);
1326 }
1327
1328
1329 /*
1330  * symlinks
1331  */
1332 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1333 {
1334         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1335         nd_set_link(nd, ci->i_symlink);
1336         return NULL;
1337 }
1338
1339 static const struct inode_operations ceph_symlink_iops = {
1340         .readlink = generic_readlink,
1341         .follow_link = ceph_sym_follow_link,
1342 };
1343
1344 /*
1345  * setattr
1346  */
1347 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1348 {
1349         struct inode *inode = dentry->d_inode;
1350         struct ceph_inode_info *ci = ceph_inode(inode);
1351         struct inode *parent_inode = dentry->d_parent->d_inode;
1352         const unsigned int ia_valid = attr->ia_valid;
1353         struct ceph_mds_request *req;
1354         struct ceph_mds_client *mdsc = &ceph_client(dentry->d_sb)->mdsc;
1355         int issued;
1356         int release = 0, dirtied = 0;
1357         int mask = 0;
1358         int err = 0;
1359         int queue_trunc = 0;
1360
1361         if (ceph_snap(inode) != CEPH_NOSNAP)
1362                 return -EROFS;
1363
1364         __ceph_do_pending_vmtruncate(inode);
1365
1366         err = inode_change_ok(inode, attr);
1367         if (err != 0)
1368                 return err;
1369
1370         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1371                                        USE_AUTH_MDS);
1372         if (IS_ERR(req))
1373                 return PTR_ERR(req);
1374
1375         spin_lock(&inode->i_lock);
1376         issued = __ceph_caps_issued(ci, NULL);
1377         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1378
1379         if (ia_valid & ATTR_UID) {
1380                 dout("setattr %p uid %d -> %d\n", inode,
1381                      inode->i_uid, attr->ia_uid);
1382                 if (issued & CEPH_CAP_AUTH_EXCL) {
1383                         inode->i_uid = attr->ia_uid;
1384                         dirtied |= CEPH_CAP_AUTH_EXCL;
1385                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1386                            attr->ia_uid != inode->i_uid) {
1387                         req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1388                         mask |= CEPH_SETATTR_UID;
1389                         release |= CEPH_CAP_AUTH_SHARED;
1390                 }
1391         }
1392         if (ia_valid & ATTR_GID) {
1393                 dout("setattr %p gid %d -> %d\n", inode,
1394                      inode->i_gid, attr->ia_gid);
1395                 if (issued & CEPH_CAP_AUTH_EXCL) {
1396                         inode->i_gid = attr->ia_gid;
1397                         dirtied |= CEPH_CAP_AUTH_EXCL;
1398                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1399                            attr->ia_gid != inode->i_gid) {
1400                         req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1401                         mask |= CEPH_SETATTR_GID;
1402                         release |= CEPH_CAP_AUTH_SHARED;
1403                 }
1404         }
1405         if (ia_valid & ATTR_MODE) {
1406                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1407                      attr->ia_mode);
1408                 if (issued & CEPH_CAP_AUTH_EXCL) {
1409                         inode->i_mode = attr->ia_mode;
1410                         dirtied |= CEPH_CAP_AUTH_EXCL;
1411                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1412                            attr->ia_mode != inode->i_mode) {
1413                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1414                         mask |= CEPH_SETATTR_MODE;
1415                         release |= CEPH_CAP_AUTH_SHARED;
1416                 }
1417         }
1418
1419         if (ia_valid & ATTR_ATIME) {
1420                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1421                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1422                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1423                 if (issued & CEPH_CAP_FILE_EXCL) {
1424                         ci->i_time_warp_seq++;
1425                         inode->i_atime = attr->ia_atime;
1426                         dirtied |= CEPH_CAP_FILE_EXCL;
1427                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1428                            timespec_compare(&inode->i_atime,
1429                                             &attr->ia_atime) < 0) {
1430                         inode->i_atime = attr->ia_atime;
1431                         dirtied |= CEPH_CAP_FILE_WR;
1432                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1433                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1434                         ceph_encode_timespec(&req->r_args.setattr.atime,
1435                                              &attr->ia_atime);
1436                         mask |= CEPH_SETATTR_ATIME;
1437                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1438                                 CEPH_CAP_FILE_WR;
1439                 }
1440         }
1441         if (ia_valid & ATTR_MTIME) {
1442                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1443                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1444                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1445                 if (issued & CEPH_CAP_FILE_EXCL) {
1446                         ci->i_time_warp_seq++;
1447                         inode->i_mtime = attr->ia_mtime;
1448                         dirtied |= CEPH_CAP_FILE_EXCL;
1449                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1450                            timespec_compare(&inode->i_mtime,
1451                                             &attr->ia_mtime) < 0) {
1452                         inode->i_mtime = attr->ia_mtime;
1453                         dirtied |= CEPH_CAP_FILE_WR;
1454                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1455                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1456                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1457                                              &attr->ia_mtime);
1458                         mask |= CEPH_SETATTR_MTIME;
1459                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1460                                 CEPH_CAP_FILE_WR;
1461                 }
1462         }
1463         if (ia_valid & ATTR_SIZE) {
1464                 dout("setattr %p size %lld -> %lld\n", inode,
1465                      inode->i_size, attr->ia_size);
1466                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1467                         err = -EINVAL;
1468                         goto out;
1469                 }
1470                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1471                     attr->ia_size > inode->i_size) {
1472                         inode->i_size = attr->ia_size;
1473                         if (attr->ia_size < inode->i_size) {
1474                                 ci->i_truncate_size = attr->ia_size;
1475                                 ci->i_truncate_pending++;
1476                                 queue_trunc = 1;
1477                         }
1478                         inode->i_blocks =
1479                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1480                         inode->i_ctime = attr->ia_ctime;
1481                         ci->i_reported_size = attr->ia_size;
1482                         dirtied |= CEPH_CAP_FILE_EXCL;
1483                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1484                            attr->ia_size != inode->i_size) {
1485                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1486                         req->r_args.setattr.old_size =
1487                                 cpu_to_le64(inode->i_size);
1488                         mask |= CEPH_SETATTR_SIZE;
1489                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1490                                 CEPH_CAP_FILE_WR;
1491                 }
1492         }
1493
1494         /* these do nothing */
1495         if (ia_valid & ATTR_CTIME) {
1496                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1497                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1498                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1499                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1500                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1501                      only ? "ctime only" : "ignored");
1502                 inode->i_ctime = attr->ia_ctime;
1503                 if (only) {
1504                         /*
1505                          * if kernel wants to dirty ctime but nothing else,
1506                          * we need to choose a cap to dirty under, or do
1507                          * a almost-no-op setattr
1508                          */
1509                         if (issued & CEPH_CAP_AUTH_EXCL)
1510                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1511                         else if (issued & CEPH_CAP_FILE_EXCL)
1512                                 dirtied |= CEPH_CAP_FILE_EXCL;
1513                         else if (issued & CEPH_CAP_XATTR_EXCL)
1514                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1515                         else
1516                                 mask |= CEPH_SETATTR_CTIME;
1517                 }
1518         }
1519         if (ia_valid & ATTR_FILE)
1520                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1521
1522         if (dirtied) {
1523                 __ceph_mark_dirty_caps(ci, dirtied);
1524                 inode->i_ctime = CURRENT_TIME;
1525         }
1526
1527         release &= issued;
1528         spin_unlock(&inode->i_lock);
1529
1530         if (queue_trunc)
1531                 __ceph_do_pending_vmtruncate(inode);
1532
1533         if (mask) {
1534                 req->r_inode = igrab(inode);
1535                 req->r_inode_drop = release;
1536                 req->r_args.setattr.mask = cpu_to_le32(mask);
1537                 req->r_num_caps = 1;
1538                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1539         }
1540         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1541              ceph_cap_string(dirtied), mask);
1542
1543         ceph_mdsc_put_request(req);
1544         __ceph_do_pending_vmtruncate(inode);
1545         return err;
1546 out:
1547         spin_unlock(&inode->i_lock);
1548         ceph_mdsc_put_request(req);
1549         return err;
1550 }
1551
1552 /*
1553  * Verify that we have a lease on the given mask.  If not,
1554  * do a getattr against an mds.
1555  */
1556 int ceph_do_getattr(struct inode *inode, int mask)
1557 {
1558         struct ceph_client *client = ceph_sb_to_client(inode->i_sb);
1559         struct ceph_mds_client *mdsc = &client->mdsc;
1560         struct ceph_mds_request *req;
1561         int err;
1562
1563         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1564                 dout("do_getattr inode %p SNAPDIR\n", inode);
1565                 return 0;
1566         }
1567
1568         dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask));
1569         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1570                 return 0;
1571
1572         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1573         if (IS_ERR(req))
1574                 return PTR_ERR(req);
1575         req->r_inode = igrab(inode);
1576         req->r_num_caps = 1;
1577         req->r_args.getattr.mask = cpu_to_le32(mask);
1578         err = ceph_mdsc_do_request(mdsc, NULL, req);
1579         ceph_mdsc_put_request(req);
1580         dout("do_getattr result=%d\n", err);
1581         return err;
1582 }
1583
1584
1585 /*
1586  * Check inode permissions.  We verify we have a valid value for
1587  * the AUTH cap, then call the generic handler.
1588  */
1589 int ceph_permission(struct inode *inode, int mask)
1590 {
1591         int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1592
1593         if (!err)
1594                 err = generic_permission(inode, mask, NULL);
1595         return err;
1596 }
1597
1598 /*
1599  * Get all attributes.  Hopefully somedata we'll have a statlite()
1600  * and can limit the fields we require to be accurate.
1601  */
1602 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1603                  struct kstat *stat)
1604 {
1605         struct inode *inode = dentry->d_inode;
1606         int err;
1607
1608         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1609         if (!err) {
1610                 generic_fillattr(inode, stat);
1611                 stat->ino = inode->i_ino;
1612                 if (ceph_snap(inode) != CEPH_NOSNAP)
1613                         stat->dev = ceph_snap(inode);
1614                 else
1615                         stat->dev = 0;
1616                 if (S_ISDIR(inode->i_mode))
1617                         stat->blksize = 65536;
1618         }
1619         return err;
1620 }