fs/ceph/xattr.c

   1 #include <linux/ceph/ceph_debug.h>
   2 #include <linux/ceph/pagelist.h>
   3
   4 #include "super.h"
   5 #include "mds_client.h"
   6
   7 #include <linux/ceph/decode.h>
   8
   9 #include <linux/xattr.h>
  10 #include <linux/posix_acl_xattr.h>
  11 #include <linux/slab.h>
  12
  13 #define XATTR_CEPH_PREFIX "ceph."
  14 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
  15
  16 static int __remove_xattr(struct ceph_inode_info *ci,
  17                           struct ceph_inode_xattr *xattr);
  18
  19 static const struct xattr_handler ceph_other_xattr_handler;
  20
  21 /*
  22  * List of handlers for synthetic system.* attributes. Other
  23  * attributes are handled directly.
  24  */
  25 const struct xattr_handler *ceph_xattr_handlers[] = {
  26 #ifdef CONFIG_CEPH_FS_POSIX_ACL
  27         &posix_acl_access_xattr_handler,
  28         &posix_acl_default_xattr_handler,
  29 #endif
  30         &ceph_other_xattr_handler,
  31         NULL,
  32 };
  33
  34 static bool ceph_is_valid_xattr(const char *name)
  35 {
  36         return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
  37                !strncmp(name, XATTR_SECURITY_PREFIX,
  38                         XATTR_SECURITY_PREFIX_LEN) ||
  39                !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
  40                !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
  41 }
  42
  43 /*
  44  * These define virtual xattrs exposing the recursive directory
  45  * statistics and layout metadata.
  46  */
  47 struct ceph_vxattr {
  48         char *name;
  49         size_t name_size;       /* strlen(name) + 1 (for '\0') */
  50         size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
  51                               size_t size);
  52         bool readonly, hidden;
  53         bool (*exists_cb)(struct ceph_inode_info *ci);
  54 };
  55
  56 /* layouts */
  57
  58 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
  59 {
  60         struct ceph_file_layout *fl = &ci->i_layout;
  61         return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
  62                 fl->object_size > 0 || fl->pool_id >= 0 ||
  63                 rcu_dereference_raw(fl->pool_ns) != NULL);
  64 }
  65
  66 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
  67                                    size_t size)
  68 {
  69         struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
  70         struct ceph_osd_client *osdc = &fsc->client->osdc;
  71         struct ceph_string *pool_ns;
  72         s64 pool = ci->i_layout.pool_id;
  73         const char *pool_name;
  74         const char *ns_field = " pool_namespace=";
  75         char buf[128];
  76         size_t len, total_len = 0;
  77         int ret;
  78
  79         pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
  80
  81         dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
  82         down_read(&osdc->lock);
  83         pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
  84         if (pool_name) {
  85                 len = snprintf(buf, sizeof(buf),
  86                 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
  87                 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
  88                 ci->i_layout.object_size);
  89                 total_len = len + strlen(pool_name);
  90         } else {
  91                 len = snprintf(buf, sizeof(buf),
  92                 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
  93                 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
  94                 ci->i_layout.object_size, (unsigned long long)pool);
  95                 total_len = len;
  96         }
  97
  98         if (pool_ns)
  99                 total_len += strlen(ns_field) + pool_ns->len;
 100
 101         if (!size) {
 102                 ret = total_len;
 103         } else if (total_len > size) {
 104                 ret = -ERANGE;
 105         } else {
 106                 memcpy(val, buf, len);
 107                 ret = len;
 108                 if (pool_name) {
 109                         len = strlen(pool_name);
 110                         memcpy(val + ret, pool_name, len);
 111                         ret += len;
 112                 }
 113                 if (pool_ns) {
 114                         len = strlen(ns_field);
 115                         memcpy(val + ret, ns_field, len);
 116                         ret += len;
 117                         memcpy(val + ret, pool_ns->str, pool_ns->len);
 118                         ret += pool_ns->len;
 119                 }
 120         }
 121         up_read(&osdc->lock);
 122         ceph_put_string(pool_ns);
 123         return ret;
 124 }
 125
 126 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
 127                                                char *val, size_t size)
 128 {
 129         return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
 130 }
 131
 132 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
 133                                                 char *val, size_t size)
 134 {
 135         return snprintf(val, size, "%u", ci->i_layout.stripe_count);
 136 }
 137
 138 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
 139                                                char *val, size_t size)
 140 {
 141         return snprintf(val, size, "%u", ci->i_layout.object_size);
 142 }
 143
 144 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
 145                                         char *val, size_t size)
 146 {
 147         int ret;
 148         struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
 149         struct ceph_osd_client *osdc = &fsc->client->osdc;
 150         s64 pool = ci->i_layout.pool_id;
 151         const char *pool_name;
 152
 153         down_read(&osdc->lock);
 154         pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
 155         if (pool_name)
 156                 ret = snprintf(val, size, "%s", pool_name);
 157         else
 158                 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
 159         up_read(&osdc->lock);
 160         return ret;
 161 }
 162
 163 static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
 164                                                   char *val, size_t size)
 165 {
 166         int ret = 0;
 167         struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
 168         if (ns) {
 169                 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
 170                 ceph_put_string(ns);
 171         }
 172         return ret;
 173 }
 174
 175 /* directories */
 176
 177 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
 178                                         size_t size)
 179 {
 180         return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
 181 }
 182
 183 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
 184                                       size_t size)
 185 {
 186         return snprintf(val, size, "%lld", ci->i_files);
 187 }
 188
 189 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
 190                                         size_t size)
 191 {
 192         return snprintf(val, size, "%lld", ci->i_subdirs);
 193 }
 194
 195 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
 196                                          size_t size)
 197 {
 198         return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
 199 }
 200
 201 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
 202                                        size_t size)
 203 {
 204         return snprintf(val, size, "%lld", ci->i_rfiles);
 205 }
 206
 207 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
 208                                          size_t size)
 209 {
 210         return snprintf(val, size, "%lld", ci->i_rsubdirs);
 211 }
 212
 213 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
 214                                        size_t size)
 215 {
 216         return snprintf(val, size, "%lld", ci->i_rbytes);
 217 }
 218
 219 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
 220                                        size_t size)
 221 {
 222         return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
 223                         (long)ci->i_rctime.tv_nsec);
 224 }
 225
 226
 227 #define CEPH_XATTR_NAME(_type, _name)   XATTR_CEPH_PREFIX #_type "." #_name
 228 #define CEPH_XATTR_NAME2(_type, _name, _name2)  \
 229         XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
 230
 231 #define XATTR_NAME_CEPH(_type, _name)                                   \
 232         {                                                               \
 233                 .name = CEPH_XATTR_NAME(_type, _name),                  \
 234                 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
 235                 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
 236                 .readonly = true,                               \
 237                 .hidden = false,                                \
 238                 .exists_cb = NULL,                      \
 239         }
 240 #define XATTR_LAYOUT_FIELD(_type, _name, _field)                        \
 241         {                                                               \
 242                 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
 243                 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
 244                 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
 245                 .readonly = false,                              \
 246                 .hidden = true,                 \
 247                 .exists_cb = ceph_vxattrcb_layout_exists,       \
 248         }
 249
 250 static struct ceph_vxattr ceph_dir_vxattrs[] = {
 251         {
 252                 .name = "ceph.dir.layout",
 253                 .name_size = sizeof("ceph.dir.layout"),
 254                 .getxattr_cb = ceph_vxattrcb_layout,
 255                 .readonly = false,
 256                 .hidden = true,
 257                 .exists_cb = ceph_vxattrcb_layout_exists,
 258         },
 259         XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
 260         XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
 261         XATTR_LAYOUT_FIELD(dir, layout, object_size),
 262         XATTR_LAYOUT_FIELD(dir, layout, pool),
 263         XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
 264         XATTR_NAME_CEPH(dir, entries),
 265         XATTR_NAME_CEPH(dir, files),
 266         XATTR_NAME_CEPH(dir, subdirs),
 267         XATTR_NAME_CEPH(dir, rentries),
 268         XATTR_NAME_CEPH(dir, rfiles),
 269         XATTR_NAME_CEPH(dir, rsubdirs),
 270         XATTR_NAME_CEPH(dir, rbytes),
 271         XATTR_NAME_CEPH(dir, rctime),
 272         { .name = NULL, 0 }     /* Required table terminator */
 273 };
 274 static size_t ceph_dir_vxattrs_name_size;       /* total size of all names */
 275
 276 /* files */
 277
 278 static struct ceph_vxattr ceph_file_vxattrs[] = {
 279         {
 280                 .name = "ceph.file.layout",
 281                 .name_size = sizeof("ceph.file.layout"),
 282                 .getxattr_cb = ceph_vxattrcb_layout,
 283                 .readonly = false,
 284                 .hidden = true,
 285                 .exists_cb = ceph_vxattrcb_layout_exists,
 286         },
 287         XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
 288         XATTR_LAYOUT_FIELD(file, layout, stripe_count),
 289         XATTR_LAYOUT_FIELD(file, layout, object_size),
 290         XATTR_LAYOUT_FIELD(file, layout, pool),
 291         XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
 292         { .name = NULL, 0 }     /* Required table terminator */
 293 };
 294 static size_t ceph_file_vxattrs_name_size;      /* total size of all names */
 295
 296 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
 297 {
 298         if (S_ISDIR(inode->i_mode))
 299                 return ceph_dir_vxattrs;
 300         else if (S_ISREG(inode->i_mode))
 301                 return ceph_file_vxattrs;
 302         return NULL;
 303 }
 304
 305 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
 306 {
 307         if (vxattrs == ceph_dir_vxattrs)
 308                 return ceph_dir_vxattrs_name_size;
 309         if (vxattrs == ceph_file_vxattrs)
 310                 return ceph_file_vxattrs_name_size;
 311         BUG_ON(vxattrs);
 312         return 0;
 313 }
 314
 315 /*
 316  * Compute the aggregate size (including terminating '\0') of all
 317  * virtual extended attribute names in the given vxattr table.
 318  */
 319 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
 320 {
 321         struct ceph_vxattr *vxattr;
 322         size_t size = 0;
 323
 324         for (vxattr = vxattrs; vxattr->name; vxattr++)
 325                 if (!vxattr->hidden)
 326                         size += vxattr->name_size;
 327
 328         return size;
 329 }
 330
 331 /* Routines called at initialization and exit time */
 332
 333 void __init ceph_xattr_init(void)
 334 {
 335         ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
 336         ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
 337 }
 338
 339 void ceph_xattr_exit(void)
 340 {
 341         ceph_dir_vxattrs_name_size = 0;
 342         ceph_file_vxattrs_name_size = 0;
 343 }
 344
 345 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
 346                                                 const char *name)
 347 {
 348         struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
 349
 350         if (vxattr) {
 351                 while (vxattr->name) {
 352                         if (!strcmp(vxattr->name, name))
 353                                 return vxattr;
 354                         vxattr++;
 355                 }
 356         }
 357
 358         return NULL;
 359 }
 360
 361 static int __set_xattr(struct ceph_inode_info *ci,
 362                            const char *name, int name_len,
 363                            const char *val, int val_len,
 364                            int flags, int update_xattr,
 365                            struct ceph_inode_xattr **newxattr)
 366 {
 367         struct rb_node **p;
 368         struct rb_node *parent = NULL;
 369         struct ceph_inode_xattr *xattr = NULL;
 370         int c;
 371         int new = 0;
 372
 373         p = &ci->i_xattrs.index.rb_node;
 374         while (*p) {
 375                 parent = *p;
 376                 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
 377                 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
 378                 if (c < 0)
 379                         p = &(*p)->rb_left;
 380                 else if (c > 0)
 381                         p = &(*p)->rb_right;
 382                 else {
 383                         if (name_len == xattr->name_len)
 384                                 break;
 385                         else if (name_len < xattr->name_len)
 386                                 p = &(*p)->rb_left;
 387                         else
 388                                 p = &(*p)->rb_right;
 389                 }
 390                 xattr = NULL;
 391         }
 392
 393         if (update_xattr) {
 394                 int err = 0;
 395
 396                 if (xattr && (flags & XATTR_CREATE))
 397                         err = -EEXIST;
 398                 else if (!xattr && (flags & XATTR_REPLACE))
 399                         err = -ENODATA;
 400                 if (err) {
 401                         kfree(name);
 402                         kfree(val);
 403                         kfree(*newxattr);
 404                         return err;
 405                 }
 406                 if (update_xattr < 0) {
 407                         if (xattr)
 408                                 __remove_xattr(ci, xattr);
 409                         kfree(name);
 410                         kfree(*newxattr);
 411                         return 0;
 412                 }
 413         }
 414
 415         if (!xattr) {
 416                 new = 1;
 417                 xattr = *newxattr;
 418                 xattr->name = name;
 419                 xattr->name_len = name_len;
 420                 xattr->should_free_name = update_xattr;
 421
 422                 ci->i_xattrs.count++;
 423                 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
 424         } else {
 425                 kfree(*newxattr);
 426                 *newxattr = NULL;
 427                 if (xattr->should_free_val)
 428                         kfree((void *)xattr->val);
 429
 430                 if (update_xattr) {
 431                         kfree((void *)name);
 432                         name = xattr->name;
 433                 }
 434                 ci->i_xattrs.names_size -= xattr->name_len;
 435                 ci->i_xattrs.vals_size -= xattr->val_len;
 436         }
 437         ci->i_xattrs.names_size += name_len;
 438         ci->i_xattrs.vals_size += val_len;
 439         if (val)
 440                 xattr->val = val;
 441         else
 442                 xattr->val = "";
 443
 444         xattr->val_len = val_len;
 445         xattr->dirty = update_xattr;
 446         xattr->should_free_val = (val && update_xattr);
 447
 448         if (new) {
 449                 rb_link_node(&xattr->node, parent, p);
 450                 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
 451                 dout("__set_xattr_val p=%p\n", p);
 452         }
 453
 454         dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
 455              ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
 456
 457         return 0;
 458 }
 459
 460 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
 461                            const char *name)
 462 {
 463         struct rb_node **p;
 464         struct rb_node *parent = NULL;
 465         struct ceph_inode_xattr *xattr = NULL;
 466         int name_len = strlen(name);
 467         int c;
 468
 469         p = &ci->i_xattrs.index.rb_node;
 470         while (*p) {
 471                 parent = *p;
 472                 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
 473                 c = strncmp(name, xattr->name, xattr->name_len);
 474                 if (c == 0 && name_len > xattr->name_len)
 475                         c = 1;
 476                 if (c < 0)
 477                         p = &(*p)->rb_left;
 478                 else if (c > 0)
 479                         p = &(*p)->rb_right;
 480                 else {
 481                         dout("__get_xattr %s: found %.*s\n", name,
 482                              xattr->val_len, xattr->val);
 483                         return xattr;
 484                 }
 485         }
 486
 487         dout("__get_xattr %s: not found\n", name);
 488
 489         return NULL;
 490 }
 491
 492 static void __free_xattr(struct ceph_inode_xattr *xattr)
 493 {
 494         BUG_ON(!xattr);
 495
 496         if (xattr->should_free_name)
 497                 kfree((void *)xattr->name);
 498         if (xattr->should_free_val)
 499                 kfree((void *)xattr->val);
 500
 501         kfree(xattr);
 502 }
 503
 504 static int __remove_xattr(struct ceph_inode_info *ci,
 505                           struct ceph_inode_xattr *xattr)
 506 {
 507         if (!xattr)
 508                 return -ENODATA;
 509
 510         rb_erase(&xattr->node, &ci->i_xattrs.index);
 511
 512         if (xattr->should_free_name)
 513                 kfree((void *)xattr->name);
 514         if (xattr->should_free_val)
 515                 kfree((void *)xattr->val);
 516
 517         ci->i_xattrs.names_size -= xattr->name_len;
 518         ci->i_xattrs.vals_size -= xattr->val_len;
 519         ci->i_xattrs.count--;
 520         kfree(xattr);
 521
 522         return 0;
 523 }
 524
 525 static char *__copy_xattr_names(struct ceph_inode_info *ci,
 526                                 char *dest)
 527 {
 528         struct rb_node *p;
 529         struct ceph_inode_xattr *xattr = NULL;
 530
 531         p = rb_first(&ci->i_xattrs.index);
 532         dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
 533
 534         while (p) {
 535                 xattr = rb_entry(p, struct ceph_inode_xattr, node);
 536                 memcpy(dest, xattr->name, xattr->name_len);
 537                 dest[xattr->name_len] = '\0';
 538
 539                 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
 540                      xattr->name_len, ci->i_xattrs.names_size);
 541
 542                 dest += xattr->name_len + 1;
 543                 p = rb_next(p);
 544         }
 545
 546         return dest;
 547 }
 548
 549 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
 550 {
 551         struct rb_node *p, *tmp;
 552         struct ceph_inode_xattr *xattr = NULL;
 553
 554         p = rb_first(&ci->i_xattrs.index);
 555
 556         dout("__ceph_destroy_xattrs p=%p\n", p);
 557
 558         while (p) {
 559                 xattr = rb_entry(p, struct ceph_inode_xattr, node);
 560                 tmp = p;
 561                 p = rb_next(tmp);
 562                 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
 563                      xattr->name_len, xattr->name);
 564                 rb_erase(tmp, &ci->i_xattrs.index);
 565
 566                 __free_xattr(xattr);
 567         }
 568
 569         ci->i_xattrs.names_size = 0;
 570         ci->i_xattrs.vals_size = 0;
 571         ci->i_xattrs.index_version = 0;
 572         ci->i_xattrs.count = 0;
 573         ci->i_xattrs.index = RB_ROOT;
 574 }
 575
 576 static int __build_xattrs(struct inode *inode)
 577         __releases(ci->i_ceph_lock)
 578         __acquires(ci->i_ceph_lock)
 579 {
 580         u32 namelen;
 581         u32 numattr = 0;
 582         void *p, *end;
 583         u32 len;
 584         const char *name, *val;
 585         struct ceph_inode_info *ci = ceph_inode(inode);
 586         int xattr_version;
 587         struct ceph_inode_xattr **xattrs = NULL;
 588         int err = 0;
 589         int i;
 590
 591         dout("__build_xattrs() len=%d\n",
 592              ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
 593
 594         if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
 595                 return 0; /* already built */
 596
 597         __ceph_destroy_xattrs(ci);
 598
 599 start:
 600         /* updated internal xattr rb tree */
 601         if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
 602                 p = ci->i_xattrs.blob->vec.iov_base;
 603                 end = p + ci->i_xattrs.blob->vec.iov_len;
 604                 ceph_decode_32_safe(&p, end, numattr, bad);
 605                 xattr_version = ci->i_xattrs.version;
 606                 spin_unlock(&ci->i_ceph_lock);
 607
 608                 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
 609                                  GFP_NOFS);
 610                 err = -ENOMEM;
 611                 if (!xattrs)
 612                         goto bad_lock;
 613
 614                 for (i = 0; i < numattr; i++) {
 615                         xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
 616                                             GFP_NOFS);
 617                         if (!xattrs[i])
 618                                 goto bad_lock;
 619                 }
 620
 621                 spin_lock(&ci->i_ceph_lock);
 622                 if (ci->i_xattrs.version != xattr_version) {
 623                         /* lost a race, retry */
 624                         for (i = 0; i < numattr; i++)
 625                                 kfree(xattrs[i]);
 626                         kfree(xattrs);
 627                         xattrs = NULL;
 628                         goto start;
 629                 }
 630                 err = -EIO;
 631                 while (numattr--) {
 632                         ceph_decode_32_safe(&p, end, len, bad);
 633                         namelen = len;
 634                         name = p;
 635                         p += len;
 636                         ceph_decode_32_safe(&p, end, len, bad);
 637                         val = p;
 638                         p += len;
 639
 640                         err = __set_xattr(ci, name, namelen, val, len,
 641                                           0, 0, &xattrs[numattr]);
 642
 643                         if (err < 0)
 644                                 goto bad;
 645                 }
 646                 kfree(xattrs);
 647         }
 648         ci->i_xattrs.index_version = ci->i_xattrs.version;
 649         ci->i_xattrs.dirty = false;
 650
 651         return err;
 652 bad_lock:
 653         spin_lock(&ci->i_ceph_lock);
 654 bad:
 655         if (xattrs) {
 656                 for (i = 0; i < numattr; i++)
 657                         kfree(xattrs[i]);
 658                 kfree(xattrs);
 659         }
 660         ci->i_xattrs.names_size = 0;
 661         return err;
 662 }
 663
 664 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
 665                                     int val_size)
 666 {
 667         /*
 668          * 4 bytes for the length, and additional 4 bytes per each xattr name,
 669          * 4 bytes per each value
 670          */
 671         int size = 4 + ci->i_xattrs.count*(4 + 4) +
 672                              ci->i_xattrs.names_size +
 673                              ci->i_xattrs.vals_size;
 674         dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
 675              ci->i_xattrs.count, ci->i_xattrs.names_size,
 676              ci->i_xattrs.vals_size);
 677
 678         if (name_size)
 679                 size += 4 + 4 + name_size + val_size;
 680
 681         return size;
 682 }
 683
 684 /*
 685  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
 686  * and swap into place.
 687  */
 688 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
 689 {
 690         struct rb_node *p;
 691         struct ceph_inode_xattr *xattr = NULL;
 692         void *dest;
 693
 694         dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
 695         if (ci->i_xattrs.dirty) {
 696                 int need = __get_required_blob_size(ci, 0, 0);
 697
 698                 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
 699
 700                 p = rb_first(&ci->i_xattrs.index);
 701                 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
 702
 703                 ceph_encode_32(&dest, ci->i_xattrs.count);
 704                 while (p) {
 705                         xattr = rb_entry(p, struct ceph_inode_xattr, node);
 706
 707                         ceph_encode_32(&dest, xattr->name_len);
 708                         memcpy(dest, xattr->name, xattr->name_len);
 709                         dest += xattr->name_len;
 710                         ceph_encode_32(&dest, xattr->val_len);
 711                         memcpy(dest, xattr->val, xattr->val_len);
 712                         dest += xattr->val_len;
 713
 714                         p = rb_next(p);
 715                 }
 716
 717                 /* adjust buffer len; it may be larger than we need */
 718                 ci->i_xattrs.prealloc_blob->vec.iov_len =
 719                         dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
 720
 721                 if (ci->i_xattrs.blob)
 722                         ceph_buffer_put(ci->i_xattrs.blob);
 723                 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
 724                 ci->i_xattrs.prealloc_blob = NULL;
 725                 ci->i_xattrs.dirty = false;
 726                 ci->i_xattrs.version++;
 727         }
 728 }
 729
 730 static inline int __get_request_mask(struct inode *in) {
 731         struct ceph_mds_request *req = current->journal_info;
 732         int mask = 0;
 733         if (req && req->r_target_inode == in) {
 734                 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
 735                     req->r_op == CEPH_MDS_OP_LOOKUPINO ||
 736                     req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
 737                     req->r_op == CEPH_MDS_OP_GETATTR) {
 738                         mask = le32_to_cpu(req->r_args.getattr.mask);
 739                 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
 740                            req->r_op == CEPH_MDS_OP_CREATE) {
 741                         mask = le32_to_cpu(req->r_args.open.mask);
 742                 }
 743         }
 744         return mask;
 745 }
 746
 747 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
 748                       size_t size)
 749 {
 750         struct ceph_inode_info *ci = ceph_inode(inode);
 751         struct ceph_inode_xattr *xattr;
 752         struct ceph_vxattr *vxattr = NULL;
 753         int req_mask;
 754         int err;
 755
 756         /* let's see if a virtual xattr was requested */
 757         vxattr = ceph_match_vxattr(inode, name);
 758         if (vxattr) {
 759                 err = -ENODATA;
 760                 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci)))
 761                         err = vxattr->getxattr_cb(ci, value, size);
 762                 return err;
 763         }
 764
 765         req_mask = __get_request_mask(inode);
 766
 767         spin_lock(&ci->i_ceph_lock);
 768         dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
 769              ci->i_xattrs.version, ci->i_xattrs.index_version);
 770
 771         if (ci->i_xattrs.version == 0 ||
 772             !((req_mask & CEPH_CAP_XATTR_SHARED) ||
 773               __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
 774                 spin_unlock(&ci->i_ceph_lock);
 775
 776                 /* security module gets xattr while filling trace */
 777                 if (current->journal_info != NULL) {
 778                         pr_warn_ratelimited("sync getxattr %p "
 779                                             "during filling trace\n", inode);
 780                         return -EBUSY;
 781                 }
 782
 783                 /* get xattrs from mds (if we don't already have them) */
 784                 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
 785                 if (err)
 786                         return err;
 787                 spin_lock(&ci->i_ceph_lock);
 788         }
 789
 790         err = __build_xattrs(inode);
 791         if (err < 0)
 792                 goto out;
 793
 794         err = -ENODATA;  /* == ENOATTR */
 795         xattr = __get_xattr(ci, name);
 796         if (!xattr)
 797                 goto out;
 798
 799         err = -ERANGE;
 800         if (size && size < xattr->val_len)
 801                 goto out;
 802
 803         err = xattr->val_len;
 804         if (size == 0)
 805                 goto out;
 806
 807         memcpy(value, xattr->val, xattr->val_len);
 808
 809         if (current->journal_info != NULL &&
 810             !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
 811                 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
 812 out:
 813         spin_unlock(&ci->i_ceph_lock);
 814         return err;
 815 }
 816
 817 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
 818 {
 819         struct inode *inode = d_inode(dentry);
 820         struct ceph_inode_info *ci = ceph_inode(inode);
 821         struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
 822         u32 vir_namelen = 0;
 823         u32 namelen;
 824         int err;
 825         u32 len;
 826         int i;
 827
 828         spin_lock(&ci->i_ceph_lock);
 829         dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
 830              ci->i_xattrs.version, ci->i_xattrs.index_version);
 831
 832         if (ci->i_xattrs.version == 0 ||
 833             !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
 834                 spin_unlock(&ci->i_ceph_lock);
 835                 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
 836                 if (err)
 837                         return err;
 838                 spin_lock(&ci->i_ceph_lock);
 839         }
 840
 841         err = __build_xattrs(inode);
 842         if (err < 0)
 843                 goto out;
 844         /*
 845          * Start with virtual dir xattr names (if any) (including
 846          * terminating '\0' characters for each).
 847          */
 848         vir_namelen = ceph_vxattrs_name_size(vxattrs);
 849
 850         /* adding 1 byte per each variable due to the null termination */
 851         namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
 852         err = -ERANGE;
 853         if (size && vir_namelen + namelen > size)
 854                 goto out;
 855
 856         err = namelen + vir_namelen;
 857         if (size == 0)
 858                 goto out;
 859
 860         names = __copy_xattr_names(ci, names);
 861
 862         /* virtual xattr names, too */
 863         err = namelen;
 864         if (vxattrs) {
 865                 for (i = 0; vxattrs[i].name; i++) {
 866                         if (!vxattrs[i].hidden &&
 867                             !(vxattrs[i].exists_cb &&
 868                               !vxattrs[i].exists_cb(ci))) {
 869                                 len = sprintf(names, "%s", vxattrs[i].name);
 870                                 names += len + 1;
 871                                 err += len + 1;
 872                         }
 873                 }
 874         }
 875
 876 out:
 877         spin_unlock(&ci->i_ceph_lock);
 878         return err;
 879 }
 880
 881 static int ceph_sync_setxattr(struct inode *inode, const char *name,
 882                               const char *value, size_t size, int flags)
 883 {
 884         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
 885         struct ceph_inode_info *ci = ceph_inode(inode);
 886         struct ceph_mds_request *req;
 887         struct ceph_mds_client *mdsc = fsc->mdsc;
 888         struct ceph_pagelist *pagelist = NULL;
 889         int op = CEPH_MDS_OP_SETXATTR;
 890         int err;
 891
 892         if (size > 0) {
 893                 /* copy value into pagelist */
 894                 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
 895                 if (!pagelist)
 896                         return -ENOMEM;
 897
 898                 ceph_pagelist_init(pagelist);
 899                 err = ceph_pagelist_append(pagelist, value, size);
 900                 if (err)
 901                         goto out;
 902         } else if (!value) {
 903                 if (flags & CEPH_XATTR_REPLACE)
 904                         op = CEPH_MDS_OP_RMXATTR;
 905                 else
 906                         flags |= CEPH_XATTR_REMOVE;
 907         }
 908
 909         dout("setxattr value=%.*s\n", (int)size, value);
 910
 911         /* do request */
 912         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 913         if (IS_ERR(req)) {
 914                 err = PTR_ERR(req);
 915                 goto out;
 916         }
 917
 918         req->r_path2 = kstrdup(name, GFP_NOFS);
 919         if (!req->r_path2) {
 920                 ceph_mdsc_put_request(req);
 921                 err = -ENOMEM;
 922                 goto out;
 923         }
 924
 925         if (op == CEPH_MDS_OP_SETXATTR) {
 926                 req->r_args.setxattr.flags = cpu_to_le32(flags);
 927                 req->r_pagelist = pagelist;
 928                 pagelist = NULL;
 929         }
 930
 931         req->r_inode = inode;
 932         ihold(inode);
 933         req->r_num_caps = 1;
 934         req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
 935
 936         dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
 937         err = ceph_mdsc_do_request(mdsc, NULL, req);
 938         ceph_mdsc_put_request(req);
 939         dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
 940
 941 out:
 942         if (pagelist)
 943                 ceph_pagelist_release(pagelist);
 944         return err;
 945 }
 946
 947 int __ceph_setxattr(struct inode *inode, const char *name,
 948                         const void *value, size_t size, int flags)
 949 {
 950         struct ceph_vxattr *vxattr;
 951         struct ceph_inode_info *ci = ceph_inode(inode);
 952         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
 953         struct ceph_cap_flush *prealloc_cf = NULL;
 954         int issued;
 955         int err;
 956         int dirty = 0;
 957         int name_len = strlen(name);
 958         int val_len = size;
 959         char *newname = NULL;
 960         char *newval = NULL;
 961         struct ceph_inode_xattr *xattr = NULL;
 962         int required_blob_size;
 963         bool lock_snap_rwsem = false;
 964
 965         if (ceph_snap(inode) != CEPH_NOSNAP)
 966                 return -EROFS;
 967
 968         vxattr = ceph_match_vxattr(inode, name);
 969         if (vxattr && vxattr->readonly)
 970                 return -EOPNOTSUPP;
 971
 972         /* pass any unhandled ceph.* xattrs through to the MDS */
 973         if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
 974                 goto do_sync_unlocked;
 975
 976         /* preallocate memory for xattr name, value, index node */
 977         err = -ENOMEM;
 978         newname = kmemdup(name, name_len + 1, GFP_NOFS);
 979         if (!newname)
 980                 goto out;
 981
 982         if (val_len) {
 983                 newval = kmemdup(value, val_len, GFP_NOFS);
 984                 if (!newval)
 985                         goto out;
 986         }
 987
 988         xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
 989         if (!xattr)
 990                 goto out;
 991
 992         prealloc_cf = ceph_alloc_cap_flush();
 993         if (!prealloc_cf)
 994                 goto out;
 995
 996         spin_lock(&ci->i_ceph_lock);
 997 retry:
 998         issued = __ceph_caps_issued(ci, NULL);
 999         if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1000                 goto do_sync;
1001
1002         if (!lock_snap_rwsem && !ci->i_head_snapc) {
1003                 lock_snap_rwsem = true;
1004                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1005                         spin_unlock(&ci->i_ceph_lock);
1006                         down_read(&mdsc->snap_rwsem);
1007                         spin_lock(&ci->i_ceph_lock);
1008                         goto retry;
1009                 }
1010         }
1011
1012         dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1013         __build_xattrs(inode);
1014
1015         required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1016
1017         if (!ci->i_xattrs.prealloc_blob ||
1018             required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1019                 struct ceph_buffer *blob;
1020
1021                 spin_unlock(&ci->i_ceph_lock);
1022                 dout(" preaallocating new blob size=%d\n", required_blob_size);
1023                 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1024                 if (!blob)
1025                         goto do_sync_unlocked;
1026                 spin_lock(&ci->i_ceph_lock);
1027                 if (ci->i_xattrs.prealloc_blob)
1028                         ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1029                 ci->i_xattrs.prealloc_blob = blob;
1030                 goto retry;
1031         }
1032
1033         err = __set_xattr(ci, newname, name_len, newval, val_len,
1034                           flags, value ? 1 : -1, &xattr);
1035
1036         if (!err) {
1037                 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1038                                                &prealloc_cf);
1039                 ci->i_xattrs.dirty = true;
1040                 inode->i_ctime = current_time(inode);
1041         }
1042
1043         spin_unlock(&ci->i_ceph_lock);
1044         if (lock_snap_rwsem)
1045                 up_read(&mdsc->snap_rwsem);
1046         if (dirty)
1047                 __mark_inode_dirty(inode, dirty);
1048         ceph_free_cap_flush(prealloc_cf);
1049         return err;
1050
1051 do_sync:
1052         spin_unlock(&ci->i_ceph_lock);
1053 do_sync_unlocked:
1054         if (lock_snap_rwsem)
1055                 up_read(&mdsc->snap_rwsem);
1056
1057         /* security module set xattr while filling trace */
1058         if (current->journal_info != NULL) {
1059                 pr_warn_ratelimited("sync setxattr %p "
1060                                     "during filling trace\n", inode);
1061                 err = -EBUSY;
1062         } else {
1063                 err = ceph_sync_setxattr(inode, name, value, size, flags);
1064         }
1065 out:
1066         ceph_free_cap_flush(prealloc_cf);
1067         kfree(newname);
1068         kfree(newval);
1069         kfree(xattr);
1070         return err;
1071 }
1072
1073 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1074                                   struct dentry *dentry, struct inode *inode,
1075                                   const char *name, void *value, size_t size)
1076 {
1077         if (!ceph_is_valid_xattr(name))
1078                 return -EOPNOTSUPP;
1079         return __ceph_getxattr(inode, name, value, size);
1080 }
1081
1082 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1083                                   struct dentry *unused, struct inode *inode,
1084                                   const char *name, const void *value,
1085                                   size_t size, int flags)
1086 {
1087         if (!ceph_is_valid_xattr(name))
1088                 return -EOPNOTSUPP;
1089         return __ceph_setxattr(inode, name, value, size, flags);
1090 }
1091
1092 static const struct xattr_handler ceph_other_xattr_handler = {
1093         .prefix = "",  /* match any name => handlers called with full name */
1094         .get = ceph_get_xattr_handler,
1095         .set = ceph_set_xattr_handler,
1096 };
1097
1098 #ifdef CONFIG_SECURITY
1099 bool ceph_security_xattr_wanted(struct inode *in)
1100 {
1101         return in->i_security != NULL;
1102 }
1103
1104 bool ceph_security_xattr_deadlock(struct inode *in)
1105 {
1106         struct ceph_inode_info *ci;
1107         bool ret;
1108         if (in->i_security == NULL)
1109                 return false;
1110         ci = ceph_inode(in);
1111         spin_lock(&ci->i_ceph_lock);
1112         ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1113               !(ci->i_xattrs.version > 0 &&
1114                 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1115         spin_unlock(&ci->i_ceph_lock);
1116         return ret;
1117 }
1118 #endif