fs/ceph/xattr.c

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