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