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