fs/ceph/xattr.c

   1 #include "ceph_debug.h"
   2 #include "super.h"
   3 #include "decode.h"
   4
   5 #include <linux/xattr.h>
   6 #include <linux/slab.h>
   7
   8 static bool ceph_is_valid_xattr(const char *name)
   9 {
  10         return !strncmp(name, XATTR_SECURITY_PREFIX,
  11                         XATTR_SECURITY_PREFIX_LEN) ||
  12                !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
  13                !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
  14 }
  15
  16 /*
  17  * These define virtual xattrs exposing the recursive directory
  18  * statistics and layout metadata.
  19  */
  20 struct ceph_vxattr_cb {
  21         bool readonly;
  22         char *name;
  23         size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
  24                               size_t size);
  25 };
  26
  27 /* directories */
  28
  29 static size_t ceph_vxattrcb_entries(struct ceph_inode_info *ci, char *val,
  30                                         size_t size)
  31 {
  32         return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
  33 }
  34
  35 static size_t ceph_vxattrcb_files(struct ceph_inode_info *ci, char *val,
  36                                       size_t size)
  37 {
  38         return snprintf(val, size, "%lld", ci->i_files);
  39 }
  40
  41 static size_t ceph_vxattrcb_subdirs(struct ceph_inode_info *ci, char *val,
  42                                         size_t size)
  43 {
  44         return snprintf(val, size, "%lld", ci->i_subdirs);
  45 }
  46
  47 static size_t ceph_vxattrcb_rentries(struct ceph_inode_info *ci, char *val,
  48                                          size_t size)
  49 {
  50         return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
  51 }
  52
  53 static size_t ceph_vxattrcb_rfiles(struct ceph_inode_info *ci, char *val,
  54                                        size_t size)
  55 {
  56         return snprintf(val, size, "%lld", ci->i_rfiles);
  57 }
  58
  59 static size_t ceph_vxattrcb_rsubdirs(struct ceph_inode_info *ci, char *val,
  60                                          size_t size)
  61 {
  62         return snprintf(val, size, "%lld", ci->i_rsubdirs);
  63 }
  64
  65 static size_t ceph_vxattrcb_rbytes(struct ceph_inode_info *ci, char *val,
  66                                        size_t size)
  67 {
  68         return snprintf(val, size, "%lld", ci->i_rbytes);
  69 }
  70
  71 static size_t ceph_vxattrcb_rctime(struct ceph_inode_info *ci, char *val,
  72                                        size_t size)
  73 {
  74         return snprintf(val, size, "%ld.%ld", (long)ci->i_rctime.tv_sec,
  75                         (long)ci->i_rctime.tv_nsec);
  76 }
  77
  78 static struct ceph_vxattr_cb ceph_dir_vxattrs[] = {
  79         { true, "user.ceph.dir.entries", ceph_vxattrcb_entries},
  80         { true, "user.ceph.dir.files", ceph_vxattrcb_files},
  81         { true, "user.ceph.dir.subdirs", ceph_vxattrcb_subdirs},
  82         { true, "user.ceph.dir.rentries", ceph_vxattrcb_rentries},
  83         { true, "user.ceph.dir.rfiles", ceph_vxattrcb_rfiles},
  84         { true, "user.ceph.dir.rsubdirs", ceph_vxattrcb_rsubdirs},
  85         { true, "user.ceph.dir.rbytes", ceph_vxattrcb_rbytes},
  86         { true, "user.ceph.dir.rctime", ceph_vxattrcb_rctime},
  87         { true, NULL, NULL }
  88 };
  89
  90 /* files */
  91
  92 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
  93                                    size_t size)
  94 {
  95         int ret;
  96
  97         ret = snprintf(val, size,
  98                 "chunk_bytes=%lld\nstripe_count=%lld\nobject_size=%lld\n",
  99                 (unsigned long long)ceph_file_layout_su(ci->i_layout),
 100                 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
 101                 (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
 102         if (ceph_file_layout_pg_preferred(ci->i_layout))
 103                 ret += snprintf(val + ret, size, "preferred_osd=%lld\n",
 104                             (unsigned long long)ceph_file_layout_pg_preferred(
 105                                     ci->i_layout));
 106         return ret;
 107 }
 108
 109 static struct ceph_vxattr_cb ceph_file_vxattrs[] = {
 110         { true, "user.ceph.layout", ceph_vxattrcb_layout},
 111         { NULL, NULL }
 112 };
 113
 114 static struct ceph_vxattr_cb *ceph_inode_vxattrs(struct inode *inode)
 115 {
 116         if (S_ISDIR(inode->i_mode))
 117                 return ceph_dir_vxattrs;
 118         else if (S_ISREG(inode->i_mode))
 119                 return ceph_file_vxattrs;
 120         return NULL;
 121 }
 122
 123 static struct ceph_vxattr_cb *ceph_match_vxattr(struct ceph_vxattr_cb *vxattr,
 124                                                 const char *name)
 125 {
 126         do {
 127                 if (strcmp(vxattr->name, name) == 0)
 128                         return vxattr;
 129                 vxattr++;
 130         } while (vxattr->name);
 131         return NULL;
 132 }
 133
 134 static int __set_xattr(struct ceph_inode_info *ci,
 135                            const char *name, int name_len,
 136                            const char *val, int val_len,
 137                            int dirty,
 138                            int should_free_name, int should_free_val,
 139                            struct ceph_inode_xattr **newxattr)
 140 {
 141         struct rb_node **p;
 142         struct rb_node *parent = NULL;
 143         struct ceph_inode_xattr *xattr = NULL;
 144         int c;
 145         int new = 0;
 146
 147         p = &ci->i_xattrs.index.rb_node;
 148         while (*p) {
 149                 parent = *p;
 150                 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
 151                 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
 152                 if (c < 0)
 153                         p = &(*p)->rb_left;
 154                 else if (c > 0)
 155                         p = &(*p)->rb_right;
 156                 else {
 157                         if (name_len == xattr->name_len)
 158                                 break;
 159                         else if (name_len < xattr->name_len)
 160                                 p = &(*p)->rb_left;
 161                         else
 162                                 p = &(*p)->rb_right;
 163                 }
 164                 xattr = NULL;
 165         }
 166
 167         if (!xattr) {
 168                 new = 1;
 169                 xattr = *newxattr;
 170                 xattr->name = name;
 171                 xattr->name_len = name_len;
 172                 xattr->should_free_name = should_free_name;
 173
 174                 ci->i_xattrs.count++;
 175                 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
 176         } else {
 177                 kfree(*newxattr);
 178                 *newxattr = NULL;
 179                 if (xattr->should_free_val)
 180                         kfree((void *)xattr->val);
 181
 182                 if (should_free_name) {
 183                         kfree((void *)name);
 184                         name = xattr->name;
 185                 }
 186                 ci->i_xattrs.names_size -= xattr->name_len;
 187                 ci->i_xattrs.vals_size -= xattr->val_len;
 188         }
 189         if (!xattr) {
 190                 pr_err("__set_xattr ENOMEM on %p %llx.%llx xattr %s=%s\n",
 191                        &ci->vfs_inode, ceph_vinop(&ci->vfs_inode), name,
 192                        xattr->val);
 193                 return -ENOMEM;
 194         }
 195         ci->i_xattrs.names_size += name_len;
 196         ci->i_xattrs.vals_size += val_len;
 197         if (val)
 198                 xattr->val = val;
 199         else
 200                 xattr->val = "";
 201
 202         xattr->val_len = val_len;
 203         xattr->dirty = dirty;
 204         xattr->should_free_val = (val && should_free_val);
 205
 206         if (new) {
 207                 rb_link_node(&xattr->node, parent, p);
 208                 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
 209                 dout("__set_xattr_val p=%p\n", p);
 210         }
 211
 212         dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
 213              ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
 214
 215         return 0;
 216 }
 217
 218 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
 219                            const char *name)
 220 {
 221         struct rb_node **p;
 222         struct rb_node *parent = NULL;
 223         struct ceph_inode_xattr *xattr = NULL;
 224         int c;
 225
 226         p = &ci->i_xattrs.index.rb_node;
 227         while (*p) {
 228                 parent = *p;
 229                 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
 230                 c = strncmp(name, xattr->name, xattr->name_len);
 231                 if (c < 0)
 232                         p = &(*p)->rb_left;
 233                 else if (c > 0)
 234                         p = &(*p)->rb_right;
 235                 else {
 236                         dout("__get_xattr %s: found %.*s\n", name,
 237                              xattr->val_len, xattr->val);
 238                         return xattr;
 239                 }
 240         }
 241
 242         dout("__get_xattr %s: not found\n", name);
 243
 244         return NULL;
 245 }
 246
 247 static void __free_xattr(struct ceph_inode_xattr *xattr)
 248 {
 249         BUG_ON(!xattr);
 250
 251         if (xattr->should_free_name)
 252                 kfree((void *)xattr->name);
 253         if (xattr->should_free_val)
 254                 kfree((void *)xattr->val);
 255
 256         kfree(xattr);
 257 }
 258
 259 static int __remove_xattr(struct ceph_inode_info *ci,
 260                           struct ceph_inode_xattr *xattr)
 261 {
 262         if (!xattr)
 263                 return -EOPNOTSUPP;
 264
 265         rb_erase(&xattr->node, &ci->i_xattrs.index);
 266
 267         if (xattr->should_free_name)
 268                 kfree((void *)xattr->name);
 269         if (xattr->should_free_val)
 270                 kfree((void *)xattr->val);
 271
 272         ci->i_xattrs.names_size -= xattr->name_len;
 273         ci->i_xattrs.vals_size -= xattr->val_len;
 274         ci->i_xattrs.count--;
 275         kfree(xattr);
 276
 277         return 0;
 278 }
 279
 280 static int __remove_xattr_by_name(struct ceph_inode_info *ci,
 281                            const char *name)
 282 {
 283         struct rb_node **p;
 284         struct ceph_inode_xattr *xattr;
 285         int err;
 286
 287         p = &ci->i_xattrs.index.rb_node;
 288         xattr = __get_xattr(ci, name);
 289         err = __remove_xattr(ci, xattr);
 290         return err;
 291 }
 292
 293 static char *__copy_xattr_names(struct ceph_inode_info *ci,
 294                                 char *dest)
 295 {
 296         struct rb_node *p;
 297         struct ceph_inode_xattr *xattr = NULL;
 298
 299         p = rb_first(&ci->i_xattrs.index);
 300         dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
 301
 302         while (p) {
 303                 xattr = rb_entry(p, struct ceph_inode_xattr, node);
 304                 memcpy(dest, xattr->name, xattr->name_len);
 305                 dest[xattr->name_len] = '\0';
 306
 307                 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
 308                      xattr->name_len, ci->i_xattrs.names_size);
 309
 310                 dest += xattr->name_len + 1;
 311                 p = rb_next(p);
 312         }
 313
 314         return dest;
 315 }
 316
 317 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
 318 {
 319         struct rb_node *p, *tmp;
 320         struct ceph_inode_xattr *xattr = NULL;
 321
 322         p = rb_first(&ci->i_xattrs.index);
 323
 324         dout("__ceph_destroy_xattrs p=%p\n", p);
 325
 326         while (p) {
 327                 xattr = rb_entry(p, struct ceph_inode_xattr, node);
 328                 tmp = p;
 329                 p = rb_next(tmp);
 330                 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
 331                      xattr->name_len, xattr->name);
 332                 rb_erase(tmp, &ci->i_xattrs.index);
 333
 334                 __free_xattr(xattr);
 335         }
 336
 337         ci->i_xattrs.names_size = 0;
 338         ci->i_xattrs.vals_size = 0;
 339         ci->i_xattrs.index_version = 0;
 340         ci->i_xattrs.count = 0;
 341         ci->i_xattrs.index = RB_ROOT;
 342 }
 343
 344 static int __build_xattrs(struct inode *inode)
 345 {
 346         u32 namelen;
 347         u32 numattr = 0;
 348         void *p, *end;
 349         u32 len;
 350         const char *name, *val;
 351         struct ceph_inode_info *ci = ceph_inode(inode);
 352         int xattr_version;
 353         struct ceph_inode_xattr **xattrs = NULL;
 354         int err = 0;
 355         int i;
 356
 357         dout("__build_xattrs() len=%d\n",
 358              ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
 359
 360         if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
 361                 return 0; /* already built */
 362
 363         __ceph_destroy_xattrs(ci);
 364
 365 start:
 366         /* updated internal xattr rb tree */
 367         if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
 368                 p = ci->i_xattrs.blob->vec.iov_base;
 369                 end = p + ci->i_xattrs.blob->vec.iov_len;
 370                 ceph_decode_32_safe(&p, end, numattr, bad);
 371                 xattr_version = ci->i_xattrs.version;
 372                 spin_unlock(&inode->i_lock);
 373
 374                 xattrs = kcalloc(numattr, sizeof(struct ceph_xattr *),
 375                                  GFP_NOFS);
 376                 err = -ENOMEM;
 377                 if (!xattrs)
 378                         goto bad_lock;
 379                 memset(xattrs, 0, numattr*sizeof(struct ceph_xattr *));
 380                 for (i = 0; i < numattr; i++) {
 381                         xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
 382                                             GFP_NOFS);
 383                         if (!xattrs[i])
 384                                 goto bad_lock;
 385                 }
 386
 387                 spin_lock(&inode->i_lock);
 388                 if (ci->i_xattrs.version != xattr_version) {
 389                         /* lost a race, retry */
 390                         for (i = 0; i < numattr; i++)
 391                                 kfree(xattrs[i]);
 392                         kfree(xattrs);
 393                         goto start;
 394                 }
 395                 err = -EIO;
 396                 while (numattr--) {
 397                         ceph_decode_32_safe(&p, end, len, bad);
 398                         namelen = len;
 399                         name = p;
 400                         p += len;
 401                         ceph_decode_32_safe(&p, end, len, bad);
 402                         val = p;
 403                         p += len;
 404
 405                         err = __set_xattr(ci, name, namelen, val, len,
 406                                           0, 0, 0, &xattrs[numattr]);
 407
 408                         if (err < 0)
 409                                 goto bad;
 410                 }
 411                 kfree(xattrs);
 412         }
 413         ci->i_xattrs.index_version = ci->i_xattrs.version;
 414         ci->i_xattrs.dirty = false;
 415
 416         return err;
 417 bad_lock:
 418         spin_lock(&inode->i_lock);
 419 bad:
 420         if (xattrs) {
 421                 for (i = 0; i < numattr; i++)
 422                         kfree(xattrs[i]);
 423                 kfree(xattrs);
 424         }
 425         ci->i_xattrs.names_size = 0;
 426         return err;
 427 }
 428
 429 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
 430                                     int val_size)
 431 {
 432         /*
 433          * 4 bytes for the length, and additional 4 bytes per each xattr name,
 434          * 4 bytes per each value
 435          */
 436         int size = 4 + ci->i_xattrs.count*(4 + 4) +
 437                              ci->i_xattrs.names_size +
 438                              ci->i_xattrs.vals_size;
 439         dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
 440              ci->i_xattrs.count, ci->i_xattrs.names_size,
 441              ci->i_xattrs.vals_size);
 442
 443         if (name_size)
 444                 size += 4 + 4 + name_size + val_size;
 445
 446         return size;
 447 }
 448
 449 /*
 450  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
 451  * and swap into place.
 452  */
 453 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
 454 {
 455         struct rb_node *p;
 456         struct ceph_inode_xattr *xattr = NULL;
 457         void *dest;
 458
 459         dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
 460         if (ci->i_xattrs.dirty) {
 461                 int need = __get_required_blob_size(ci, 0, 0);
 462
 463                 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
 464
 465                 p = rb_first(&ci->i_xattrs.index);
 466                 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
 467
 468                 ceph_encode_32(&dest, ci->i_xattrs.count);
 469                 while (p) {
 470                         xattr = rb_entry(p, struct ceph_inode_xattr, node);
 471
 472                         ceph_encode_32(&dest, xattr->name_len);
 473                         memcpy(dest, xattr->name, xattr->name_len);
 474                         dest += xattr->name_len;
 475                         ceph_encode_32(&dest, xattr->val_len);
 476                         memcpy(dest, xattr->val, xattr->val_len);
 477                         dest += xattr->val_len;
 478
 479                         p = rb_next(p);
 480                 }
 481
 482                 /* adjust buffer len; it may be larger than we need */
 483                 ci->i_xattrs.prealloc_blob->vec.iov_len =
 484                         dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
 485
 486                 if (ci->i_xattrs.blob)
 487                         ceph_buffer_put(ci->i_xattrs.blob);
 488                 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
 489                 ci->i_xattrs.prealloc_blob = NULL;
 490                 ci->i_xattrs.dirty = false;
 491         }
 492 }
 493
 494 ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
 495                       size_t size)
 496 {
 497         struct inode *inode = dentry->d_inode;
 498         struct ceph_inode_info *ci = ceph_inode(inode);
 499         struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
 500         int err;
 501         struct ceph_inode_xattr *xattr;
 502         struct ceph_vxattr_cb *vxattr = NULL;
 503
 504         if (!ceph_is_valid_xattr(name))
 505                 return -ENODATA;
 506
 507         /* let's see if a virtual xattr was requested */
 508         if (vxattrs)
 509                 vxattr = ceph_match_vxattr(vxattrs, name);
 510
 511         spin_lock(&inode->i_lock);
 512         dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
 513              ci->i_xattrs.version, ci->i_xattrs.index_version);
 514
 515         if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
 516             (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
 517                 goto get_xattr;
 518         } else {
 519                 spin_unlock(&inode->i_lock);
 520                 /* get xattrs from mds (if we don't already have them) */
 521                 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
 522                 if (err)
 523                         return err;
 524         }
 525
 526         spin_lock(&inode->i_lock);
 527
 528         if (vxattr && vxattr->readonly) {
 529                 err = vxattr->getxattr_cb(ci, value, size);
 530                 goto out;
 531         }
 532
 533         err = __build_xattrs(inode);
 534         if (err < 0)
 535                 goto out;
 536
 537 get_xattr:
 538         err = -ENODATA;  /* == ENOATTR */
 539         xattr = __get_xattr(ci, name);
 540         if (!xattr) {
 541                 if (vxattr)
 542                         err = vxattr->getxattr_cb(ci, value, size);
 543                 goto out;
 544         }
 545
 546         err = -ERANGE;
 547         if (size && size < xattr->val_len)
 548                 goto out;
 549
 550         err = xattr->val_len;
 551         if (size == 0)
 552                 goto out;
 553
 554         memcpy(value, xattr->val, xattr->val_len);
 555
 556 out:
 557         spin_unlock(&inode->i_lock);
 558         return err;
 559 }
 560
 561 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
 562 {
 563         struct inode *inode = dentry->d_inode;
 564         struct ceph_inode_info *ci = ceph_inode(inode);
 565         struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
 566         u32 vir_namelen = 0;
 567         u32 namelen;
 568         int err;
 569         u32 len;
 570         int i;
 571
 572         spin_lock(&inode->i_lock);
 573         dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
 574              ci->i_xattrs.version, ci->i_xattrs.index_version);
 575
 576         if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
 577             (ci->i_xattrs.index_version > ci->i_xattrs.version)) {
 578                 goto list_xattr;
 579         } else {
 580                 spin_unlock(&inode->i_lock);
 581                 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
 582                 if (err)
 583                         return err;
 584         }
 585
 586         spin_lock(&inode->i_lock);
 587
 588         err = __build_xattrs(inode);
 589         if (err < 0)
 590                 goto out;
 591
 592 list_xattr:
 593         vir_namelen = 0;
 594         /* include virtual dir xattrs */
 595         if (vxattrs)
 596                 for (i = 0; vxattrs[i].name; i++)
 597                         vir_namelen += strlen(vxattrs[i].name) + 1;
 598         /* adding 1 byte per each variable due to the null termination */
 599         namelen = vir_namelen + ci->i_xattrs.names_size + ci->i_xattrs.count;
 600         err = -ERANGE;
 601         if (size && namelen > size)
 602                 goto out;
 603
 604         err = namelen;
 605         if (size == 0)
 606                 goto out;
 607
 608         names = __copy_xattr_names(ci, names);
 609
 610         /* virtual xattr names, too */
 611         if (vxattrs)
 612                 for (i = 0; vxattrs[i].name; i++) {
 613                         len = sprintf(names, "%s", vxattrs[i].name);
 614                         names += len + 1;
 615                 }
 616
 617 out:
 618         spin_unlock(&inode->i_lock);
 619         return err;
 620 }
 621
 622 static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
 623                               const char *value, size_t size, int flags)
 624 {
 625         struct ceph_client *client = ceph_client(dentry->d_sb);
 626         struct inode *inode = dentry->d_inode;
 627         struct ceph_inode_info *ci = ceph_inode(inode);
 628         struct inode *parent_inode = dentry->d_parent->d_inode;
 629         struct ceph_mds_request *req;
 630         struct ceph_mds_client *mdsc = &client->mdsc;
 631         int err;
 632         int i, nr_pages;
 633         struct page **pages = NULL;
 634         void *kaddr;
 635
 636         /* copy value into some pages */
 637         nr_pages = calc_pages_for(0, size);
 638         if (nr_pages) {
 639                 pages = kmalloc(sizeof(pages[0])*nr_pages, GFP_NOFS);
 640                 if (!pages)
 641                         return -ENOMEM;
 642                 err = -ENOMEM;
 643                 for (i = 0; i < nr_pages; i++) {
 644                         pages[i] = alloc_page(GFP_NOFS);
 645                         if (!pages[i]) {
 646                                 nr_pages = i;
 647                                 goto out;
 648                         }
 649                         kaddr = kmap(pages[i]);
 650                         memcpy(kaddr, value + i*PAGE_CACHE_SIZE,
 651                                min(PAGE_CACHE_SIZE, size-i*PAGE_CACHE_SIZE));
 652                 }
 653         }
 654
 655         dout("setxattr value=%.*s\n", (int)size, value);
 656
 657         /* do request */
 658         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
 659                                        USE_AUTH_MDS);
 660         if (IS_ERR(req)) {
 661                 err = PTR_ERR(req);
 662                 goto out;
 663         }
 664         req->r_inode = igrab(inode);
 665         req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
 666         req->r_num_caps = 1;
 667         req->r_args.setxattr.flags = cpu_to_le32(flags);
 668         req->r_path2 = kstrdup(name, GFP_NOFS);
 669
 670         req->r_pages = pages;
 671         req->r_num_pages = nr_pages;
 672         req->r_data_len = size;
 673
 674         dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
 675         err = ceph_mdsc_do_request(mdsc, parent_inode, req);
 676         ceph_mdsc_put_request(req);
 677         dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
 678
 679 out:
 680         if (pages) {
 681                 for (i = 0; i < nr_pages; i++)
 682                         __free_page(pages[i]);
 683                 kfree(pages);
 684         }
 685         return err;
 686 }
 687
 688 int ceph_setxattr(struct dentry *dentry, const char *name,
 689                   const void *value, size_t size, int flags)
 690 {
 691         struct inode *inode = dentry->d_inode;
 692         struct ceph_inode_info *ci = ceph_inode(inode);
 693         struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
 694         int err;
 695         int name_len = strlen(name);
 696         int val_len = size;
 697         char *newname = NULL;
 698         char *newval = NULL;
 699         struct ceph_inode_xattr *xattr = NULL;
 700         int issued;
 701         int required_blob_size;
 702
 703         if (ceph_snap(inode) != CEPH_NOSNAP)
 704                 return -EROFS;
 705
 706         if (!ceph_is_valid_xattr(name))
 707                 return -EOPNOTSUPP;
 708
 709         if (vxattrs) {
 710                 struct ceph_vxattr_cb *vxattr =
 711                         ceph_match_vxattr(vxattrs, name);
 712                 if (vxattr && vxattr->readonly)
 713                         return -EOPNOTSUPP;
 714         }
 715
 716         /* preallocate memory for xattr name, value, index node */
 717         err = -ENOMEM;
 718         newname = kmalloc(name_len + 1, GFP_NOFS);
 719         if (!newname)
 720                 goto out;
 721         memcpy(newname, name, name_len + 1);
 722
 723         if (val_len) {
 724                 newval = kmalloc(val_len + 1, GFP_NOFS);
 725                 if (!newval)
 726                         goto out;
 727                 memcpy(newval, value, val_len);
 728                 newval[val_len] = '\0';
 729         }
 730
 731         xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
 732         if (!xattr)
 733                 goto out;
 734
 735         spin_lock(&inode->i_lock);
 736 retry:
 737         issued = __ceph_caps_issued(ci, NULL);
 738         if (!(issued & CEPH_CAP_XATTR_EXCL))
 739                 goto do_sync;
 740         __build_xattrs(inode);
 741
 742         required_blob_size = __get_required_blob_size(ci, name_len, val_len);
 743
 744         if (!ci->i_xattrs.prealloc_blob ||
 745             required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
 746                 struct ceph_buffer *blob = NULL;
 747
 748                 spin_unlock(&inode->i_lock);
 749                 dout(" preaallocating new blob size=%d\n", required_blob_size);
 750                 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
 751                 if (!blob)
 752                         goto out;
 753                 spin_lock(&inode->i_lock);
 754                 if (ci->i_xattrs.prealloc_blob)
 755                         ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 756                 ci->i_xattrs.prealloc_blob = blob;
 757                 goto retry;
 758         }
 759
 760         dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
 761         err = __set_xattr(ci, newname, name_len, newval,
 762                           val_len, 1, 1, 1, &xattr);
 763         __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
 764         ci->i_xattrs.dirty = true;
 765         inode->i_ctime = CURRENT_TIME;
 766         spin_unlock(&inode->i_lock);
 767
 768         return err;
 769
 770 do_sync:
 771         spin_unlock(&inode->i_lock);
 772         err = ceph_sync_setxattr(dentry, name, value, size, flags);
 773 out:
 774         kfree(newname);
 775         kfree(newval);
 776         kfree(xattr);
 777         return err;
 778 }
 779
 780 static int ceph_send_removexattr(struct dentry *dentry, const char *name)
 781 {
 782         struct ceph_client *client = ceph_client(dentry->d_sb);
 783         struct ceph_mds_client *mdsc = &client->mdsc;
 784         struct inode *inode = dentry->d_inode;
 785         struct inode *parent_inode = dentry->d_parent->d_inode;
 786         struct ceph_mds_request *req;
 787         int err;
 788
 789         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
 790                                        USE_AUTH_MDS);
 791         if (IS_ERR(req))
 792                 return PTR_ERR(req);
 793         req->r_inode = igrab(inode);
 794         req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
 795         req->r_num_caps = 1;
 796         req->r_path2 = kstrdup(name, GFP_NOFS);
 797
 798         err = ceph_mdsc_do_request(mdsc, parent_inode, req);
 799         ceph_mdsc_put_request(req);
 800         return err;
 801 }
 802
 803 int ceph_removexattr(struct dentry *dentry, const char *name)
 804 {
 805         struct inode *inode = dentry->d_inode;
 806         struct ceph_inode_info *ci = ceph_inode(inode);
 807         struct ceph_vxattr_cb *vxattrs = ceph_inode_vxattrs(inode);
 808         int issued;
 809         int err;
 810
 811         if (ceph_snap(inode) != CEPH_NOSNAP)
 812                 return -EROFS;
 813
 814         if (!ceph_is_valid_xattr(name))
 815                 return -EOPNOTSUPP;
 816
 817         if (vxattrs) {
 818                 struct ceph_vxattr_cb *vxattr =
 819                         ceph_match_vxattr(vxattrs, name);
 820                 if (vxattr && vxattr->readonly)
 821                         return -EOPNOTSUPP;
 822         }
 823
 824         spin_lock(&inode->i_lock);
 825         __build_xattrs(inode);
 826         issued = __ceph_caps_issued(ci, NULL);
 827         dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
 828
 829         if (!(issued & CEPH_CAP_XATTR_EXCL))
 830                 goto do_sync;
 831
 832         err = __remove_xattr_by_name(ceph_inode(inode), name);
 833         __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
 834         ci->i_xattrs.dirty = true;
 835         inode->i_ctime = CURRENT_TIME;
 836
 837         spin_unlock(&inode->i_lock);
 838
 839         return err;
 840 do_sync:
 841         spin_unlock(&inode->i_lock);
 842         err = ceph_send_removexattr(dentry, name);
 843         return err;
 844 }
 845