fs/ceph/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3 #include <linux/ceph/pagelist.h>
   4
   5 #include "super.h"
   6 #include "mds_client.h"
   7
   8 #include <linux/ceph/decode.h>
   9
  10 #include <linux/xattr.h>
  11 #include <linux/posix_acl_xattr.h>
  12 #include <linux/slab.h>
  13
  14 #define XATTR_CEPH_PREFIX "ceph."
  15 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
  16
  17 static int __remove_xattr(struct ceph_inode_info *ci,
  18                           struct ceph_inode_xattr *xattr);
  19
  20 static const struct xattr_handler ceph_other_xattr_handler;
  21
  22 /*
  23  * List of handlers for synthetic system.* attributes. Other
  24  * attributes are handled directly.
  25  */
  26 const struct xattr_handler *ceph_xattr_handlers[] = {
  27 #ifdef CONFIG_CEPH_FS_POSIX_ACL
  28         &posix_acl_access_xattr_handler,
  29         &posix_acl_default_xattr_handler,
  30 #endif
  31         &ceph_other_xattr_handler,
  32         NULL,
  33 };
  34
  35 static bool ceph_is_valid_xattr(const char *name)
  36 {
  37         return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
  38                !strncmp(name, XATTR_SECURITY_PREFIX,
  39                         XATTR_SECURITY_PREFIX_LEN) ||
  40                !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
  41                !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
  42 }
  43
  44 /*
  45  * These define virtual xattrs exposing the recursive directory
  46  * statistics and layout metadata.
  47  */
  48 struct ceph_vxattr {
  49         char *name;
  50         size_t name_size;       /* strlen(name) + 1 (for '\0') */
  51         size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
  52                               size_t size);
  53         bool readonly, hidden;
  54         bool (*exists_cb)(struct ceph_inode_info *ci);
  55 };
  56
  57 /* layouts */
  58
  59 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
  60 {
  61         struct ceph_file_layout *fl = &ci->i_layout;
  62         return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
  63                 fl->object_size > 0 || fl->pool_id >= 0 ||
  64                 rcu_dereference_raw(fl->pool_ns) != NULL);
  65 }
  66
  67 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
  68                                    size_t size)
  69 {
  70         struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
  71         struct ceph_osd_client *osdc = &fsc->client->osdc;
  72         struct ceph_string *pool_ns;
  73         s64 pool = ci->i_layout.pool_id;
  74         const char *pool_name;
  75         const char *ns_field = " pool_namespace=";
  76         char buf[128];
  77         size_t len, total_len = 0;
  78         ssize_t ret;
  79
  80         pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
  81
  82         dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
  83         down_read(&osdc->lock);
  84         pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
  85         if (pool_name) {
  86                 len = snprintf(buf, sizeof(buf),
  87                 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
  88                 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
  89                 ci->i_layout.object_size);
  90                 total_len = len + strlen(pool_name);
  91         } else {
  92                 len = snprintf(buf, sizeof(buf),
  93                 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
  94                 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
  95                 ci->i_layout.object_size, (unsigned long long)pool);
  96                 total_len = len;
  97         }
  98
  99         if (pool_ns)
 100                 total_len += strlen(ns_field) + pool_ns->len;
 101
 102         ret = total_len;
 103         if (size >= total_len) {
 104                 memcpy(val, buf, len);
 105                 ret = len;
 106                 if (pool_name) {
 107                         len = strlen(pool_name);
 108                         memcpy(val + ret, pool_name, len);
 109                         ret += len;
 110                 }
 111                 if (pool_ns) {
 112                         len = strlen(ns_field);
 113                         memcpy(val + ret, ns_field, len);
 114                         ret += len;
 115                         memcpy(val + ret, pool_ns->str, pool_ns->len);
 116                         ret += pool_ns->len;
 117                 }
 118         }
 119         up_read(&osdc->lock);
 120         ceph_put_string(pool_ns);
 121         return ret;
 122 }
 123
 124 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
 125                                                char *val, size_t size)
 126 {
 127         return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
 128 }
 129
 130 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
 131                                                 char *val, size_t size)
 132 {
 133         return snprintf(val, size, "%u", ci->i_layout.stripe_count);
 134 }
 135
 136 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
 137                                                char *val, size_t size)
 138 {
 139         return snprintf(val, size, "%u", ci->i_layout.object_size);
 140 }
 141
 142 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
 143                                         char *val, size_t size)
 144 {
 145         int ret;
 146         struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
 147         struct ceph_osd_client *osdc = &fsc->client->osdc;
 148         s64 pool = ci->i_layout.pool_id;
 149         const char *pool_name;
 150
 151         down_read(&osdc->lock);
 152         pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
 153         if (pool_name)
 154                 ret = snprintf(val, size, "%s", pool_name);
 155         else
 156                 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
 157         up_read(&osdc->lock);
 158         return ret;
 159 }
 160
 161 static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
 162                                                   char *val, size_t size)
 163 {
 164         int ret = 0;
 165         struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
 166         if (ns) {
 167                 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
 168                 ceph_put_string(ns);
 169         }
 170         return ret;
 171 }
 172
 173 /* directories */
 174
 175 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
 176                                         size_t size)
 177 {
 178         return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
 179 }
 180
 181 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
 182                                       size_t size)
 183 {
 184         return snprintf(val, size, "%lld", ci->i_files);
 185 }
 186
 187 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
 188                                         size_t size)
 189 {
 190         return snprintf(val, size, "%lld", ci->i_subdirs);
 191 }
 192
 193 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
 194                                          size_t size)
 195 {
 196         return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
 197 }
 198
 199 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
 200                                        size_t size)
 201 {
 202         return snprintf(val, size, "%lld", ci->i_rfiles);
 203 }
 204
 205 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
 206                                          size_t size)
 207 {
 208         return snprintf(val, size, "%lld", ci->i_rsubdirs);
 209 }
 210
 211 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
 212                                        size_t size)
 213 {
 214         return snprintf(val, size, "%lld", ci->i_rbytes);
 215 }
 216
 217 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
 218                                        size_t size)
 219 {
 220         return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
 221                         (long)ci->i_rctime.tv_nsec);
 222 }
 223
 224 /* quotas */
 225
 226 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
 227 {
 228         return (ci->i_max_files || ci->i_max_bytes);
 229 }
 230
 231 static size_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
 232                                   size_t size)
 233 {
 234         return snprintf(val, size, "max_bytes=%llu max_files=%llu",
 235                         ci->i_max_bytes, ci->i_max_files);
 236 }
 237
 238 static size_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
 239                                             char *val, size_t size)
 240 {
 241         return snprintf(val, size, "%llu", ci->i_max_bytes);
 242 }
 243
 244 static size_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
 245                                             char *val, size_t size)
 246 {
 247         return snprintf(val, size, "%llu", ci->i_max_files);
 248 }
 249
 250 #define CEPH_XATTR_NAME(_type, _name)   XATTR_CEPH_PREFIX #_type "." #_name
 251 #define CEPH_XATTR_NAME2(_type, _name, _name2)  \
 252         XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
 253
 254 #define XATTR_NAME_CEPH(_type, _name)                                   \
 255         {                                                               \
 256                 .name = CEPH_XATTR_NAME(_type, _name),                  \
 257                 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
 258                 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
 259                 .readonly = true,                               \
 260                 .hidden = false,                                \
 261                 .exists_cb = NULL,                      \
 262         }
 263 #define XATTR_LAYOUT_FIELD(_type, _name, _field)                        \
 264         {                                                               \
 265                 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
 266                 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
 267                 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
 268                 .readonly = false,                              \
 269                 .hidden = true,                 \
 270                 .exists_cb = ceph_vxattrcb_layout_exists,       \
 271         }
 272 #define XATTR_QUOTA_FIELD(_type, _name)                                 \
 273         {                                                               \
 274                 .name = CEPH_XATTR_NAME(_type, _name),                  \
 275                 .name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),     \
 276                 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,   \
 277                 .readonly = false,                                      \
 278                 .hidden = true,                                         \
 279                 .exists_cb = ceph_vxattrcb_quota_exists,                \
 280         }
 281
 282 static struct ceph_vxattr ceph_dir_vxattrs[] = {
 283         {
 284                 .name = "ceph.dir.layout",
 285                 .name_size = sizeof("ceph.dir.layout"),
 286                 .getxattr_cb = ceph_vxattrcb_layout,
 287                 .readonly = false,
 288                 .hidden = true,
 289                 .exists_cb = ceph_vxattrcb_layout_exists,
 290         },
 291         XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
 292         XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
 293         XATTR_LAYOUT_FIELD(dir, layout, object_size),
 294         XATTR_LAYOUT_FIELD(dir, layout, pool),
 295         XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
 296         XATTR_NAME_CEPH(dir, entries),
 297         XATTR_NAME_CEPH(dir, files),
 298         XATTR_NAME_CEPH(dir, subdirs),
 299         XATTR_NAME_CEPH(dir, rentries),
 300         XATTR_NAME_CEPH(dir, rfiles),
 301         XATTR_NAME_CEPH(dir, rsubdirs),
 302         XATTR_NAME_CEPH(dir, rbytes),
 303         XATTR_NAME_CEPH(dir, rctime),
 304         {
 305                 .name = "ceph.quota",
 306                 .name_size = sizeof("ceph.quota"),
 307                 .getxattr_cb = ceph_vxattrcb_quota,
 308                 .readonly = false,
 309                 .hidden = true,
 310                 .exists_cb = ceph_vxattrcb_quota_exists,
 311         },
 312         XATTR_QUOTA_FIELD(quota, max_bytes),
 313         XATTR_QUOTA_FIELD(quota, max_files),
 314         { .name = NULL, 0 }     /* Required table terminator */
 315 };
 316 static size_t ceph_dir_vxattrs_name_size;       /* total size of all names */
 317
 318 /* files */
 319
 320 static struct ceph_vxattr ceph_file_vxattrs[] = {
 321         {
 322                 .name = "ceph.file.layout",
 323                 .name_size = sizeof("ceph.file.layout"),
 324                 .getxattr_cb = ceph_vxattrcb_layout,
 325                 .readonly = false,
 326                 .hidden = true,
 327                 .exists_cb = ceph_vxattrcb_layout_exists,
 328         },
 329         XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
 330         XATTR_LAYOUT_FIELD(file, layout, stripe_count),
 331         XATTR_LAYOUT_FIELD(file, layout, object_size),
 332         XATTR_LAYOUT_FIELD(file, layout, pool),
 333         XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
 334         { .name = NULL, 0 }     /* Required table terminator */
 335 };
 336 static size_t ceph_file_vxattrs_name_size;      /* total size of all names */
 337
 338 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
 339 {
 340         if (S_ISDIR(inode->i_mode))
 341                 return ceph_dir_vxattrs;
 342         else if (S_ISREG(inode->i_mode))
 343                 return ceph_file_vxattrs;
 344         return NULL;
 345 }
 346
 347 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
 348 {
 349         if (vxattrs == ceph_dir_vxattrs)
 350                 return ceph_dir_vxattrs_name_size;
 351         if (vxattrs == ceph_file_vxattrs)
 352                 return ceph_file_vxattrs_name_size;
 353         BUG_ON(vxattrs);
 354         return 0;
 355 }
 356
 357 /*
 358  * Compute the aggregate size (including terminating '\0') of all
 359  * virtual extended attribute names in the given vxattr table.
 360  */
 361 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
 362 {
 363         struct ceph_vxattr *vxattr;
 364         size_t size = 0;
 365
 366         for (vxattr = vxattrs; vxattr->name; vxattr++)
 367                 if (!vxattr->hidden)
 368                         size += vxattr->name_size;
 369
 370         return size;
 371 }
 372
 373 /* Routines called at initialization and exit time */
 374
 375 void __init ceph_xattr_init(void)
 376 {
 377         ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
 378         ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
 379 }
 380
 381 void ceph_xattr_exit(void)
 382 {
 383         ceph_dir_vxattrs_name_size = 0;
 384         ceph_file_vxattrs_name_size = 0;
 385 }
 386
 387 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
 388                                                 const char *name)
 389 {
 390         struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
 391
 392         if (vxattr) {
 393                 while (vxattr->name) {
 394                         if (!strcmp(vxattr->name, name))
 395                                 return vxattr;
 396                         vxattr++;
 397                 }
 398         }
 399
 400         return NULL;
 401 }
 402
 403 static int __set_xattr(struct ceph_inode_info *ci,
 404                            const char *name, int name_len,
 405                            const char *val, int val_len,
 406                            int flags, int update_xattr,
 407                            struct ceph_inode_xattr **newxattr)
 408 {
 409         struct rb_node **p;
 410         struct rb_node *parent = NULL;
 411         struct ceph_inode_xattr *xattr = NULL;
 412         int c;
 413         int new = 0;
 414
 415         p = &ci->i_xattrs.index.rb_node;
 416         while (*p) {
 417                 parent = *p;
 418                 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
 419                 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
 420                 if (c < 0)
 421                         p = &(*p)->rb_left;
 422                 else if (c > 0)
 423                         p = &(*p)->rb_right;
 424                 else {
 425                         if (name_len == xattr->name_len)
 426                                 break;
 427                         else if (name_len < xattr->name_len)
 428                                 p = &(*p)->rb_left;
 429                         else
 430                                 p = &(*p)->rb_right;
 431                 }
 432                 xattr = NULL;
 433         }
 434
 435         if (update_xattr) {
 436                 int err = 0;
 437
 438                 if (xattr && (flags & XATTR_CREATE))
 439                         err = -EEXIST;
 440                 else if (!xattr && (flags & XATTR_REPLACE))
 441                         err = -ENODATA;
 442                 if (err) {
 443                         kfree(name);
 444                         kfree(val);
 445                         kfree(*newxattr);
 446                         return err;
 447                 }
 448                 if (update_xattr < 0) {
 449                         if (xattr)
 450                                 __remove_xattr(ci, xattr);
 451                         kfree(name);
 452                         kfree(*newxattr);
 453                         return 0;
 454                 }
 455         }
 456
 457         if (!xattr) {
 458                 new = 1;
 459                 xattr = *newxattr;
 460                 xattr->name = name;
 461                 xattr->name_len = name_len;
 462                 xattr->should_free_name = update_xattr;
 463
 464                 ci->i_xattrs.count++;
 465                 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
 466         } else {
 467                 kfree(*newxattr);
 468                 *newxattr = NULL;
 469                 if (xattr->should_free_val)
 470                         kfree((void *)xattr->val);
 471
 472                 if (update_xattr) {
 473                         kfree((void *)name);
 474                         name = xattr->name;
 475                 }
 476                 ci->i_xattrs.names_size -= xattr->name_len;
 477                 ci->i_xattrs.vals_size -= xattr->val_len;
 478         }
 479         ci->i_xattrs.names_size += name_len;
 480         ci->i_xattrs.vals_size += val_len;
 481         if (val)
 482                 xattr->val = val;
 483         else
 484                 xattr->val = "";
 485
 486         xattr->val_len = val_len;
 487         xattr->dirty = update_xattr;
 488         xattr->should_free_val = (val && update_xattr);
 489
 490         if (new) {
 491                 rb_link_node(&xattr->node, parent, p);
 492                 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
 493                 dout("__set_xattr_val p=%p\n", p);
 494         }
 495
 496         dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
 497              ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
 498
 499         return 0;
 500 }
 501
 502 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
 503                            const char *name)
 504 {
 505         struct rb_node **p;
 506         struct rb_node *parent = NULL;
 507         struct ceph_inode_xattr *xattr = NULL;
 508         int name_len = strlen(name);
 509         int c;
 510
 511         p = &ci->i_xattrs.index.rb_node;
 512         while (*p) {
 513                 parent = *p;
 514                 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
 515                 c = strncmp(name, xattr->name, xattr->name_len);
 516                 if (c == 0 && name_len > xattr->name_len)
 517                         c = 1;
 518                 if (c < 0)
 519                         p = &(*p)->rb_left;
 520                 else if (c > 0)
 521                         p = &(*p)->rb_right;
 522                 else {
 523                         dout("__get_xattr %s: found %.*s\n", name,
 524                              xattr->val_len, xattr->val);
 525                         return xattr;
 526                 }
 527         }
 528
 529         dout("__get_xattr %s: not found\n", name);
 530
 531         return NULL;
 532 }
 533
 534 static void __free_xattr(struct ceph_inode_xattr *xattr)
 535 {
 536         BUG_ON(!xattr);
 537
 538         if (xattr->should_free_name)
 539                 kfree((void *)xattr->name);
 540         if (xattr->should_free_val)
 541                 kfree((void *)xattr->val);
 542
 543         kfree(xattr);
 544 }
 545
 546 static int __remove_xattr(struct ceph_inode_info *ci,
 547                           struct ceph_inode_xattr *xattr)
 548 {
 549         if (!xattr)
 550                 return -ENODATA;
 551
 552         rb_erase(&xattr->node, &ci->i_xattrs.index);
 553
 554         if (xattr->should_free_name)
 555                 kfree((void *)xattr->name);
 556         if (xattr->should_free_val)
 557                 kfree((void *)xattr->val);
 558
 559         ci->i_xattrs.names_size -= xattr->name_len;
 560         ci->i_xattrs.vals_size -= xattr->val_len;
 561         ci->i_xattrs.count--;
 562         kfree(xattr);
 563
 564         return 0;
 565 }
 566
 567 static char *__copy_xattr_names(struct ceph_inode_info *ci,
 568                                 char *dest)
 569 {
 570         struct rb_node *p;
 571         struct ceph_inode_xattr *xattr = NULL;
 572
 573         p = rb_first(&ci->i_xattrs.index);
 574         dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
 575
 576         while (p) {
 577                 xattr = rb_entry(p, struct ceph_inode_xattr, node);
 578                 memcpy(dest, xattr->name, xattr->name_len);
 579                 dest[xattr->name_len] = '\0';
 580
 581                 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
 582                      xattr->name_len, ci->i_xattrs.names_size);
 583
 584                 dest += xattr->name_len + 1;
 585                 p = rb_next(p);
 586         }
 587
 588         return dest;
 589 }
 590
 591 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
 592 {
 593         struct rb_node *p, *tmp;
 594         struct ceph_inode_xattr *xattr = NULL;
 595
 596         p = rb_first(&ci->i_xattrs.index);
 597
 598         dout("__ceph_destroy_xattrs p=%p\n", p);
 599
 600         while (p) {
 601                 xattr = rb_entry(p, struct ceph_inode_xattr, node);
 602                 tmp = p;
 603                 p = rb_next(tmp);
 604                 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
 605                      xattr->name_len, xattr->name);
 606                 rb_erase(tmp, &ci->i_xattrs.index);
 607
 608                 __free_xattr(xattr);
 609         }
 610
 611         ci->i_xattrs.names_size = 0;
 612         ci->i_xattrs.vals_size = 0;
 613         ci->i_xattrs.index_version = 0;
 614         ci->i_xattrs.count = 0;
 615         ci->i_xattrs.index = RB_ROOT;
 616 }
 617
 618 static int __build_xattrs(struct inode *inode)
 619         __releases(ci->i_ceph_lock)
 620         __acquires(ci->i_ceph_lock)
 621 {
 622         u32 namelen;
 623         u32 numattr = 0;
 624         void *p, *end;
 625         u32 len;
 626         const char *name, *val;
 627         struct ceph_inode_info *ci = ceph_inode(inode);
 628         int xattr_version;
 629         struct ceph_inode_xattr **xattrs = NULL;
 630         int err = 0;
 631         int i;
 632
 633         dout("__build_xattrs() len=%d\n",
 634              ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
 635
 636         if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
 637                 return 0; /* already built */
 638
 639         __ceph_destroy_xattrs(ci);
 640
 641 start:
 642         /* updated internal xattr rb tree */
 643         if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
 644                 p = ci->i_xattrs.blob->vec.iov_base;
 645                 end = p + ci->i_xattrs.blob->vec.iov_len;
 646                 ceph_decode_32_safe(&p, end, numattr, bad);
 647                 xattr_version = ci->i_xattrs.version;
 648                 spin_unlock(&ci->i_ceph_lock);
 649
 650                 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
 651                                  GFP_NOFS);
 652                 err = -ENOMEM;
 653                 if (!xattrs)
 654                         goto bad_lock;
 655
 656                 for (i = 0; i < numattr; i++) {
 657                         xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
 658                                             GFP_NOFS);
 659                         if (!xattrs[i])
 660                                 goto bad_lock;
 661                 }
 662
 663                 spin_lock(&ci->i_ceph_lock);
 664                 if (ci->i_xattrs.version != xattr_version) {
 665                         /* lost a race, retry */
 666                         for (i = 0; i < numattr; i++)
 667                                 kfree(xattrs[i]);
 668                         kfree(xattrs);
 669                         xattrs = NULL;
 670                         goto start;
 671                 }
 672                 err = -EIO;
 673                 while (numattr--) {
 674                         ceph_decode_32_safe(&p, end, len, bad);
 675                         namelen = len;
 676                         name = p;
 677                         p += len;
 678                         ceph_decode_32_safe(&p, end, len, bad);
 679                         val = p;
 680                         p += len;
 681
 682                         err = __set_xattr(ci, name, namelen, val, len,
 683                                           0, 0, &xattrs[numattr]);
 684
 685                         if (err < 0)
 686                                 goto bad;
 687                 }
 688                 kfree(xattrs);
 689         }
 690         ci->i_xattrs.index_version = ci->i_xattrs.version;
 691         ci->i_xattrs.dirty = false;
 692
 693         return err;
 694 bad_lock:
 695         spin_lock(&ci->i_ceph_lock);
 696 bad:
 697         if (xattrs) {
 698                 for (i = 0; i < numattr; i++)
 699                         kfree(xattrs[i]);
 700                 kfree(xattrs);
 701         }
 702         ci->i_xattrs.names_size = 0;
 703         return err;
 704 }
 705
 706 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
 707                                     int val_size)
 708 {
 709         /*
 710          * 4 bytes for the length, and additional 4 bytes per each xattr name,
 711          * 4 bytes per each value
 712          */
 713         int size = 4 + ci->i_xattrs.count*(4 + 4) +
 714                              ci->i_xattrs.names_size +
 715                              ci->i_xattrs.vals_size;
 716         dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
 717              ci->i_xattrs.count, ci->i_xattrs.names_size,
 718              ci->i_xattrs.vals_size);
 719
 720         if (name_size)
 721                 size += 4 + 4 + name_size + val_size;
 722
 723         return size;
 724 }
 725
 726 /*
 727  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
 728  * and swap into place.  It returns the old i_xattrs.blob (or NULL) so
 729  * that it can be freed by the caller as the i_ceph_lock is likely to be
 730  * held.
 731  */
 732 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
 733 {
 734         struct rb_node *p;
 735         struct ceph_inode_xattr *xattr = NULL;
 736         struct ceph_buffer *old_blob = NULL;
 737         void *dest;
 738
 739         dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
 740         if (ci->i_xattrs.dirty) {
 741                 int need = __get_required_blob_size(ci, 0, 0);
 742
 743                 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
 744
 745                 p = rb_first(&ci->i_xattrs.index);
 746                 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
 747
 748                 ceph_encode_32(&dest, ci->i_xattrs.count);
 749                 while (p) {
 750                         xattr = rb_entry(p, struct ceph_inode_xattr, node);
 751
 752                         ceph_encode_32(&dest, xattr->name_len);
 753                         memcpy(dest, xattr->name, xattr->name_len);
 754                         dest += xattr->name_len;
 755                         ceph_encode_32(&dest, xattr->val_len);
 756                         memcpy(dest, xattr->val, xattr->val_len);
 757                         dest += xattr->val_len;
 758
 759                         p = rb_next(p);
 760                 }
 761
 762                 /* adjust buffer len; it may be larger than we need */
 763                 ci->i_xattrs.prealloc_blob->vec.iov_len =
 764                         dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
 765
 766                 if (ci->i_xattrs.blob)
 767                         old_blob = ci->i_xattrs.blob;
 768                 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
 769                 ci->i_xattrs.prealloc_blob = NULL;
 770                 ci->i_xattrs.dirty = false;
 771                 ci->i_xattrs.version++;
 772         }
 773
 774         return old_blob;
 775 }
 776
 777 static inline int __get_request_mask(struct inode *in) {
 778         struct ceph_mds_request *req = current->journal_info;
 779         int mask = 0;
 780         if (req && req->r_target_inode == in) {
 781                 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
 782                     req->r_op == CEPH_MDS_OP_LOOKUPINO ||
 783                     req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
 784                     req->r_op == CEPH_MDS_OP_GETATTR) {
 785                         mask = le32_to_cpu(req->r_args.getattr.mask);
 786                 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
 787                            req->r_op == CEPH_MDS_OP_CREATE) {
 788                         mask = le32_to_cpu(req->r_args.open.mask);
 789                 }
 790         }
 791         return mask;
 792 }
 793
 794 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
 795                       size_t size)
 796 {
 797         struct ceph_inode_info *ci = ceph_inode(inode);
 798         struct ceph_inode_xattr *xattr;
 799         struct ceph_vxattr *vxattr = NULL;
 800         int req_mask;
 801         int err;
 802
 803         /* let's see if a virtual xattr was requested */
 804         vxattr = ceph_match_vxattr(inode, name);
 805         if (vxattr) {
 806                 err = ceph_do_getattr(inode, 0, true);
 807                 if (err)
 808                         return err;
 809                 err = -ENODATA;
 810                 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
 811                         err = vxattr->getxattr_cb(ci, value, size);
 812                         if (size && size < err)
 813                                 err = -ERANGE;
 814                 }
 815                 return err;
 816         }
 817
 818         req_mask = __get_request_mask(inode);
 819
 820         spin_lock(&ci->i_ceph_lock);
 821         dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
 822              ci->i_xattrs.version, ci->i_xattrs.index_version);
 823
 824         if (ci->i_xattrs.version == 0 ||
 825             !((req_mask & CEPH_CAP_XATTR_SHARED) ||
 826               __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
 827                 spin_unlock(&ci->i_ceph_lock);
 828
 829                 /* security module gets xattr while filling trace */
 830                 if (current->journal_info) {
 831                         pr_warn_ratelimited("sync getxattr %p "
 832                                             "during filling trace\n", inode);
 833                         return -EBUSY;
 834                 }
 835
 836                 /* get xattrs from mds (if we don't already have them) */
 837                 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
 838                 if (err)
 839                         return err;
 840                 spin_lock(&ci->i_ceph_lock);
 841         }
 842
 843         err = __build_xattrs(inode);
 844         if (err < 0)
 845                 goto out;
 846
 847         err = -ENODATA;  /* == ENOATTR */
 848         xattr = __get_xattr(ci, name);
 849         if (!xattr)
 850                 goto out;
 851
 852         err = -ERANGE;
 853         if (size && size < xattr->val_len)
 854                 goto out;
 855
 856         err = xattr->val_len;
 857         if (size == 0)
 858                 goto out;
 859
 860         memcpy(value, xattr->val, xattr->val_len);
 861
 862         if (current->journal_info &&
 863             !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
 864                 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
 865 out:
 866         spin_unlock(&ci->i_ceph_lock);
 867         return err;
 868 }
 869
 870 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
 871 {
 872         struct inode *inode = d_inode(dentry);
 873         struct ceph_inode_info *ci = ceph_inode(inode);
 874         struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
 875         u32 vir_namelen = 0;
 876         u32 namelen;
 877         int err;
 878         u32 len;
 879         int i;
 880
 881         spin_lock(&ci->i_ceph_lock);
 882         dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
 883              ci->i_xattrs.version, ci->i_xattrs.index_version);
 884
 885         if (ci->i_xattrs.version == 0 ||
 886             !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
 887                 spin_unlock(&ci->i_ceph_lock);
 888                 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
 889                 if (err)
 890                         return err;
 891                 spin_lock(&ci->i_ceph_lock);
 892         }
 893
 894         err = __build_xattrs(inode);
 895         if (err < 0)
 896                 goto out;
 897         /*
 898          * Start with virtual dir xattr names (if any) (including
 899          * terminating '\0' characters for each).
 900          */
 901         vir_namelen = ceph_vxattrs_name_size(vxattrs);
 902
 903         /* adding 1 byte per each variable due to the null termination */
 904         namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
 905         err = -ERANGE;
 906         if (size && vir_namelen + namelen > size)
 907                 goto out;
 908
 909         err = namelen + vir_namelen;
 910         if (size == 0)
 911                 goto out;
 912
 913         names = __copy_xattr_names(ci, names);
 914
 915         /* virtual xattr names, too */
 916         err = namelen;
 917         if (vxattrs) {
 918                 for (i = 0; vxattrs[i].name; i++) {
 919                         if (!vxattrs[i].hidden &&
 920                             !(vxattrs[i].exists_cb &&
 921                               !vxattrs[i].exists_cb(ci))) {
 922                                 len = sprintf(names, "%s", vxattrs[i].name);
 923                                 names += len + 1;
 924                                 err += len + 1;
 925                         }
 926                 }
 927         }
 928
 929 out:
 930         spin_unlock(&ci->i_ceph_lock);
 931         return err;
 932 }
 933
 934 static int ceph_sync_setxattr(struct inode *inode, const char *name,
 935                               const char *value, size_t size, int flags)
 936 {
 937         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
 938         struct ceph_inode_info *ci = ceph_inode(inode);
 939         struct ceph_mds_request *req;
 940         struct ceph_mds_client *mdsc = fsc->mdsc;
 941         struct ceph_pagelist *pagelist = NULL;
 942         int op = CEPH_MDS_OP_SETXATTR;
 943         int err;
 944
 945         if (size > 0) {
 946                 /* copy value into pagelist */
 947                 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
 948                 if (!pagelist)
 949                         return -ENOMEM;
 950
 951                 ceph_pagelist_init(pagelist);
 952                 err = ceph_pagelist_append(pagelist, value, size);
 953                 if (err)
 954                         goto out;
 955         } else if (!value) {
 956                 if (flags & CEPH_XATTR_REPLACE)
 957                         op = CEPH_MDS_OP_RMXATTR;
 958                 else
 959                         flags |= CEPH_XATTR_REMOVE;
 960         }
 961
 962         dout("setxattr value=%.*s\n", (int)size, value);
 963
 964         /* do request */
 965         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 966         if (IS_ERR(req)) {
 967                 err = PTR_ERR(req);
 968                 goto out;
 969         }
 970
 971         req->r_path2 = kstrdup(name, GFP_NOFS);
 972         if (!req->r_path2) {
 973                 ceph_mdsc_put_request(req);
 974                 err = -ENOMEM;
 975                 goto out;
 976         }
 977
 978         if (op == CEPH_MDS_OP_SETXATTR) {
 979                 req->r_args.setxattr.flags = cpu_to_le32(flags);
 980                 req->r_pagelist = pagelist;
 981                 pagelist = NULL;
 982         }
 983
 984         req->r_inode = inode;
 985         ihold(inode);
 986         req->r_num_caps = 1;
 987         req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
 988
 989         dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
 990         err = ceph_mdsc_do_request(mdsc, NULL, req);
 991         ceph_mdsc_put_request(req);
 992         dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
 993
 994 out:
 995         if (pagelist)
 996                 ceph_pagelist_release(pagelist);
 997         return err;
 998 }
 999
1000 int __ceph_setxattr(struct inode *inode, const char *name,
1001                         const void *value, size_t size, int flags)
1002 {
1003         struct ceph_vxattr *vxattr;
1004         struct ceph_inode_info *ci = ceph_inode(inode);
1005         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1006         struct ceph_cap_flush *prealloc_cf = NULL;
1007         struct ceph_buffer *old_blob = NULL;
1008         int issued;
1009         int err;
1010         int dirty = 0;
1011         int name_len = strlen(name);
1012         int val_len = size;
1013         char *newname = NULL;
1014         char *newval = NULL;
1015         struct ceph_inode_xattr *xattr = NULL;
1016         int required_blob_size;
1017         bool lock_snap_rwsem = false;
1018
1019         if (ceph_snap(inode) != CEPH_NOSNAP)
1020                 return -EROFS;
1021
1022         vxattr = ceph_match_vxattr(inode, name);
1023         if (vxattr && vxattr->readonly)
1024                 return -EOPNOTSUPP;
1025
1026         /* pass any unhandled ceph.* xattrs through to the MDS */
1027         if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1028                 goto do_sync_unlocked;
1029
1030         /* preallocate memory for xattr name, value, index node */
1031         err = -ENOMEM;
1032         newname = kmemdup(name, name_len + 1, GFP_NOFS);
1033         if (!newname)
1034                 goto out;
1035
1036         if (val_len) {
1037                 newval = kmemdup(value, val_len, GFP_NOFS);
1038                 if (!newval)
1039                         goto out;
1040         }
1041
1042         xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1043         if (!xattr)
1044                 goto out;
1045
1046         prealloc_cf = ceph_alloc_cap_flush();
1047         if (!prealloc_cf)
1048                 goto out;
1049
1050         spin_lock(&ci->i_ceph_lock);
1051 retry:
1052         issued = __ceph_caps_issued(ci, NULL);
1053         if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1054                 goto do_sync;
1055
1056         if (!lock_snap_rwsem && !ci->i_head_snapc) {
1057                 lock_snap_rwsem = true;
1058                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1059                         spin_unlock(&ci->i_ceph_lock);
1060                         down_read(&mdsc->snap_rwsem);
1061                         spin_lock(&ci->i_ceph_lock);
1062                         goto retry;
1063                 }
1064         }
1065
1066         dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1067         __build_xattrs(inode);
1068
1069         required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1070
1071         if (!ci->i_xattrs.prealloc_blob ||
1072             required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1073                 struct ceph_buffer *blob;
1074
1075                 spin_unlock(&ci->i_ceph_lock);
1076                 ceph_buffer_put(old_blob); /* Shouldn't be required */
1077                 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1078                 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1079                 if (!blob)
1080                         goto do_sync_unlocked;
1081                 spin_lock(&ci->i_ceph_lock);
1082                 /* prealloc_blob can't be released while holding i_ceph_lock */
1083                 if (ci->i_xattrs.prealloc_blob)
1084                         old_blob = ci->i_xattrs.prealloc_blob;
1085                 ci->i_xattrs.prealloc_blob = blob;
1086                 goto retry;
1087         }
1088
1089         err = __set_xattr(ci, newname, name_len, newval, val_len,
1090                           flags, value ? 1 : -1, &xattr);
1091
1092         if (!err) {
1093                 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1094                                                &prealloc_cf);
1095                 ci->i_xattrs.dirty = true;
1096                 inode->i_ctime = current_time(inode);
1097         }
1098
1099         spin_unlock(&ci->i_ceph_lock);
1100         ceph_buffer_put(old_blob);
1101         if (lock_snap_rwsem)
1102                 up_read(&mdsc->snap_rwsem);
1103         if (dirty)
1104                 __mark_inode_dirty(inode, dirty);
1105         ceph_free_cap_flush(prealloc_cf);
1106         return err;
1107
1108 do_sync:
1109         spin_unlock(&ci->i_ceph_lock);
1110 do_sync_unlocked:
1111         if (lock_snap_rwsem)
1112                 up_read(&mdsc->snap_rwsem);
1113
1114         /* security module set xattr while filling trace */
1115         if (current->journal_info) {
1116                 pr_warn_ratelimited("sync setxattr %p "
1117                                     "during filling trace\n", inode);
1118                 err = -EBUSY;
1119         } else {
1120                 err = ceph_sync_setxattr(inode, name, value, size, flags);
1121         }
1122 out:
1123         ceph_free_cap_flush(prealloc_cf);
1124         kfree(newname);
1125         kfree(newval);
1126         kfree(xattr);
1127         return err;
1128 }
1129
1130 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1131                                   struct dentry *dentry, struct inode *inode,
1132                                   const char *name, void *value, size_t size)
1133 {
1134         if (!ceph_is_valid_xattr(name))
1135                 return -EOPNOTSUPP;
1136         return __ceph_getxattr(inode, name, value, size);
1137 }
1138
1139 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1140                                   struct dentry *unused, struct inode *inode,
1141                                   const char *name, const void *value,
1142                                   size_t size, int flags)
1143 {
1144         if (!ceph_is_valid_xattr(name))
1145                 return -EOPNOTSUPP;
1146         return __ceph_setxattr(inode, name, value, size, flags);
1147 }
1148
1149 static const struct xattr_handler ceph_other_xattr_handler = {
1150         .prefix = "",  /* match any name => handlers called with full name */
1151         .get = ceph_get_xattr_handler,
1152         .set = ceph_set_xattr_handler,
1153 };
1154
1155 #ifdef CONFIG_SECURITY
1156 bool ceph_security_xattr_wanted(struct inode *in)
1157 {
1158         return in->i_security != NULL;
1159 }
1160
1161 bool ceph_security_xattr_deadlock(struct inode *in)
1162 {
1163         struct ceph_inode_info *ci;
1164         bool ret;
1165         if (!in->i_security)
1166                 return false;
1167         ci = ceph_inode(in);
1168         spin_lock(&ci->i_ceph_lock);
1169         ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1170               !(ci->i_xattrs.version > 0 &&
1171                 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1172         spin_unlock(&ci->i_ceph_lock);
1173         return ret;
1174 }
1175 #endif