ceph/src/spdk/dpdk/lib/librte_mbuf/rte_mbuf.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation.
   3  * Copyright 2014 6WIND S.A.
   4  */
   5
   6 #include <string.h>
   7 #include <stdio.h>
   8 #include <stdlib.h>
   9 #include <stdint.h>
  10 #include <stdarg.h>
  11 #include <inttypes.h>
  12 #include <errno.h>
  13 #include <ctype.h>
  14 #include <sys/queue.h>
  15
  16 #include <rte_compat.h>
  17 #include <rte_debug.h>
  18 #include <rte_common.h>
  19 #include <rte_log.h>
  20 #include <rte_memory.h>
  21 #include <rte_launch.h>
  22 #include <rte_eal.h>
  23 #include <rte_per_lcore.h>
  24 #include <rte_lcore.h>
  25 #include <rte_atomic.h>
  26 #include <rte_branch_prediction.h>
  27 #include <rte_mempool.h>
  28 #include <rte_mbuf.h>
  29 #include <rte_mbuf_pool_ops.h>
  30 #include <rte_string_fns.h>
  31 #include <rte_hexdump.h>
  32 #include <rte_errno.h>
  33 #include <rte_memcpy.h>
  34
  35 /*
  36  * pktmbuf pool constructor, given as a callback function to
  37  * rte_mempool_create(), or called directly if using
  38  * rte_mempool_create_empty()/rte_mempool_populate()
  39  */
  40 void
  41 rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg)
  42 {
  43         struct rte_pktmbuf_pool_private *user_mbp_priv, *mbp_priv;
  44         struct rte_pktmbuf_pool_private default_mbp_priv;
  45         uint16_t roomsz;
  46
  47         RTE_ASSERT(mp->elt_size >= sizeof(struct rte_mbuf));
  48
  49         /* if no structure is provided, assume no mbuf private area */
  50         user_mbp_priv = opaque_arg;
  51         if (user_mbp_priv == NULL) {
  52                 memset(&default_mbp_priv, 0, sizeof(default_mbp_priv));
  53                 if (mp->elt_size > sizeof(struct rte_mbuf))
  54                         roomsz = mp->elt_size - sizeof(struct rte_mbuf);
  55                 else
  56                         roomsz = 0;
  57                 default_mbp_priv.mbuf_data_room_size = roomsz;
  58                 user_mbp_priv = &default_mbp_priv;
  59         }
  60
  61         RTE_ASSERT(mp->elt_size >= sizeof(struct rte_mbuf) +
  62                 ((user_mbp_priv->flags & RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF) ?
  63                         sizeof(struct rte_mbuf_ext_shared_info) :
  64                         user_mbp_priv->mbuf_data_room_size) +
  65                 user_mbp_priv->mbuf_priv_size);
  66         RTE_ASSERT((user_mbp_priv->flags &
  67                     ~RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF) == 0);
  68
  69         mbp_priv = rte_mempool_get_priv(mp);
  70         memcpy(mbp_priv, user_mbp_priv, sizeof(*mbp_priv));
  71 }
  72
  73 /*
  74  * pktmbuf constructor, given as a callback function to
  75  * rte_mempool_obj_iter() or rte_mempool_create().
  76  * Set the fields of a packet mbuf to their default values.
  77  */
  78 void
  79 rte_pktmbuf_init(struct rte_mempool *mp,
  80                  __rte_unused void *opaque_arg,
  81                  void *_m,
  82                  __rte_unused unsigned i)
  83 {
  84         struct rte_mbuf *m = _m;
  85         uint32_t mbuf_size, buf_len, priv_size;
  86
  87         priv_size = rte_pktmbuf_priv_size(mp);
  88         mbuf_size = sizeof(struct rte_mbuf) + priv_size;
  89         buf_len = rte_pktmbuf_data_room_size(mp);
  90
  91         RTE_ASSERT(RTE_ALIGN(priv_size, RTE_MBUF_PRIV_ALIGN) == priv_size);
  92         RTE_ASSERT(mp->elt_size >= mbuf_size);
  93         RTE_ASSERT(buf_len <= UINT16_MAX);
  94
  95         memset(m, 0, mbuf_size);
  96         /* start of buffer is after mbuf structure and priv data */
  97         m->priv_size = priv_size;
  98         m->buf_addr = (char *)m + mbuf_size;
  99         m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 100         m->buf_len = (uint16_t)buf_len;
 101
 102         /* keep some headroom between start of buffer and data */
 103         m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, (uint16_t)m->buf_len);
 104
 105         /* init some constant fields */
 106         m->pool = mp;
 107         m->nb_segs = 1;
 108         m->port = MBUF_INVALID_PORT;
 109         rte_mbuf_refcnt_set(m, 1);
 110         m->next = NULL;
 111 }
 112
 113 /*
 114  * @internal The callback routine called when reference counter in shinfo
 115  * for mbufs with pinned external buffer reaches zero. It means there is
 116  * no more reference to buffer backing mbuf and this one should be freed.
 117  * This routine is called for the regular (not with pinned external or
 118  * indirect buffer) mbufs on detaching from the mbuf with pinned external
 119  * buffer.
 120  */
 121 static void
 122 rte_pktmbuf_free_pinned_extmem(void *addr, void *opaque)
 123 {
 124         struct rte_mbuf *m = opaque;
 125
 126         RTE_SET_USED(addr);
 127         RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
 128         RTE_ASSERT(RTE_MBUF_HAS_PINNED_EXTBUF(m));
 129         RTE_ASSERT(m->shinfo->fcb_opaque == m);
 130
 131         rte_mbuf_ext_refcnt_set(m->shinfo, 1);
 132         m->ol_flags = EXT_ATTACHED_MBUF;
 133         if (m->next != NULL) {
 134                 m->next = NULL;
 135                 m->nb_segs = 1;
 136         }
 137         rte_mbuf_raw_free(m);
 138 }
 139
 140 /** The context to initialize the mbufs with pinned external buffers. */
 141 struct rte_pktmbuf_extmem_init_ctx {
 142         const struct rte_pktmbuf_extmem *ext_mem; /* descriptor array. */
 143         unsigned int ext_num; /* number of descriptors in array. */
 144         unsigned int ext; /* loop descriptor index. */
 145         size_t off; /* loop buffer offset. */
 146 };
 147
 148 /**
 149  * @internal Packet mbuf constructor for pools with pinned external memory.
 150  *
 151  * This function initializes some fields in the mbuf structure that are
 152  * not modified by the user once created (origin pool, buffer start
 153  * address, and so on). This function is given as a callback function to
 154  * rte_mempool_obj_iter() called from rte_mempool_create_extmem().
 155  *
 156  * @param mp
 157  *   The mempool from which mbufs originate.
 158  * @param opaque_arg
 159  *   A pointer to the rte_pktmbuf_extmem_init_ctx - initialization
 160  *   context structure
 161  * @param m
 162  *   The mbuf to initialize.
 163  * @param i
 164  *   The index of the mbuf in the pool table.
 165  */
 166 static void
 167 __rte_pktmbuf_init_extmem(struct rte_mempool *mp,
 168                           void *opaque_arg,
 169                           void *_m,
 170                           __rte_unused unsigned int i)
 171 {
 172         struct rte_mbuf *m = _m;
 173         struct rte_pktmbuf_extmem_init_ctx *ctx = opaque_arg;
 174         const struct rte_pktmbuf_extmem *ext_mem;
 175         uint32_t mbuf_size, buf_len, priv_size;
 176         struct rte_mbuf_ext_shared_info *shinfo;
 177
 178         priv_size = rte_pktmbuf_priv_size(mp);
 179         mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 180         buf_len = rte_pktmbuf_data_room_size(mp);
 181
 182         RTE_ASSERT(RTE_ALIGN(priv_size, RTE_MBUF_PRIV_ALIGN) == priv_size);
 183         RTE_ASSERT(mp->elt_size >= mbuf_size);
 184         RTE_ASSERT(buf_len <= UINT16_MAX);
 185
 186         memset(m, 0, mbuf_size);
 187         m->priv_size = priv_size;
 188         m->buf_len = (uint16_t)buf_len;
 189
 190         /* set the data buffer pointers to external memory */
 191         ext_mem = ctx->ext_mem + ctx->ext;
 192
 193         RTE_ASSERT(ctx->ext < ctx->ext_num);
 194         RTE_ASSERT(ctx->off < ext_mem->buf_len);
 195
 196         m->buf_addr = RTE_PTR_ADD(ext_mem->buf_ptr, ctx->off);
 197         m->buf_iova = ext_mem->buf_iova == RTE_BAD_IOVA ?
 198                       RTE_BAD_IOVA : (ext_mem->buf_iova + ctx->off);
 199
 200         ctx->off += ext_mem->elt_size;
 201         if (ctx->off >= ext_mem->buf_len) {
 202                 ctx->off = 0;
 203                 ++ctx->ext;
 204         }
 205         /* keep some headroom between start of buffer and data */
 206         m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, (uint16_t)m->buf_len);
 207
 208         /* init some constant fields */
 209         m->pool = mp;
 210         m->nb_segs = 1;
 211         m->port = MBUF_INVALID_PORT;
 212         m->ol_flags = EXT_ATTACHED_MBUF;
 213         rte_mbuf_refcnt_set(m, 1);
 214         m->next = NULL;
 215
 216         /* init external buffer shared info items */
 217         shinfo = RTE_PTR_ADD(m, mbuf_size);
 218         m->shinfo = shinfo;
 219         shinfo->free_cb = rte_pktmbuf_free_pinned_extmem;
 220         shinfo->fcb_opaque = m;
 221         rte_mbuf_ext_refcnt_set(shinfo, 1);
 222 }
 223
 224 /* Helper to create a mbuf pool with given mempool ops name*/
 225 struct rte_mempool *
 226 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
 227         unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
 228         int socket_id, const char *ops_name)
 229 {
 230         struct rte_mempool *mp;
 231         struct rte_pktmbuf_pool_private mbp_priv;
 232         const char *mp_ops_name = ops_name;
 233         unsigned elt_size;
 234         int ret;
 235
 236         if (RTE_ALIGN(priv_size, RTE_MBUF_PRIV_ALIGN) != priv_size) {
 237                 RTE_LOG(ERR, MBUF, "mbuf priv_size=%u is not aligned\n",
 238                         priv_size);
 239                 rte_errno = EINVAL;
 240                 return NULL;
 241         }
 242         elt_size = sizeof(struct rte_mbuf) + (unsigned)priv_size +
 243                 (unsigned)data_room_size;
 244         memset(&mbp_priv, 0, sizeof(mbp_priv));
 245         mbp_priv.mbuf_data_room_size = data_room_size;
 246         mbp_priv.mbuf_priv_size = priv_size;
 247
 248         mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
 249                  sizeof(struct rte_pktmbuf_pool_private), socket_id, 0);
 250         if (mp == NULL)
 251                 return NULL;
 252
 253         if (mp_ops_name == NULL)
 254                 mp_ops_name = rte_mbuf_best_mempool_ops();
 255         ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
 256         if (ret != 0) {
 257                 RTE_LOG(ERR, MBUF, "error setting mempool handler\n");
 258                 rte_mempool_free(mp);
 259                 rte_errno = -ret;
 260                 return NULL;
 261         }
 262         rte_pktmbuf_pool_init(mp, &mbp_priv);
 263
 264         ret = rte_mempool_populate_default(mp);
 265         if (ret < 0) {
 266                 rte_mempool_free(mp);
 267                 rte_errno = -ret;
 268                 return NULL;
 269         }
 270
 271         rte_mempool_obj_iter(mp, rte_pktmbuf_init, NULL);
 272
 273         return mp;
 274 }
 275
 276 /* helper to create a mbuf pool */
 277 struct rte_mempool *
 278 rte_pktmbuf_pool_create(const char *name, unsigned int n,
 279         unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
 280         int socket_id)
 281 {
 282         return rte_pktmbuf_pool_create_by_ops(name, n, cache_size, priv_size,
 283                         data_room_size, socket_id, NULL);
 284 }
 285
 286 /* Helper to create a mbuf pool with pinned external data buffers. */
 287 struct rte_mempool *
 288 rte_pktmbuf_pool_create_extbuf(const char *name, unsigned int n,
 289         unsigned int cache_size, uint16_t priv_size,
 290         uint16_t data_room_size, int socket_id,
 291         const struct rte_pktmbuf_extmem *ext_mem,
 292         unsigned int ext_num)
 293 {
 294         struct rte_mempool *mp;
 295         struct rte_pktmbuf_pool_private mbp_priv;
 296         struct rte_pktmbuf_extmem_init_ctx init_ctx;
 297         const char *mp_ops_name;
 298         unsigned int elt_size;
 299         unsigned int i, n_elts = 0;
 300         int ret;
 301
 302         if (RTE_ALIGN(priv_size, RTE_MBUF_PRIV_ALIGN) != priv_size) {
 303                 RTE_LOG(ERR, MBUF, "mbuf priv_size=%u is not aligned\n",
 304                         priv_size);
 305                 rte_errno = EINVAL;
 306                 return NULL;
 307         }
 308         /* Check the external memory descriptors. */
 309         for (i = 0; i < ext_num; i++) {
 310                 const struct rte_pktmbuf_extmem *extm = ext_mem + i;
 311
 312                 if (!extm->elt_size || !extm->buf_len || !extm->buf_ptr) {
 313                         RTE_LOG(ERR, MBUF, "invalid extmem descriptor\n");
 314                         rte_errno = EINVAL;
 315                         return NULL;
 316                 }
 317                 if (data_room_size > extm->elt_size) {
 318                         RTE_LOG(ERR, MBUF, "ext elt_size=%u is too small\n",
 319                                 priv_size);
 320                         rte_errno = EINVAL;
 321                         return NULL;
 322                 }
 323                 n_elts += extm->buf_len / extm->elt_size;
 324         }
 325         /* Check whether enough external memory provided. */
 326         if (n_elts < n) {
 327                 RTE_LOG(ERR, MBUF, "not enough extmem\n");
 328                 rte_errno = ENOMEM;
 329                 return NULL;
 330         }
 331         elt_size = sizeof(struct rte_mbuf) +
 332                    (unsigned int)priv_size +
 333                    sizeof(struct rte_mbuf_ext_shared_info);
 334
 335         memset(&mbp_priv, 0, sizeof(mbp_priv));
 336         mbp_priv.mbuf_data_room_size = data_room_size;
 337         mbp_priv.mbuf_priv_size = priv_size;
 338         mbp_priv.flags = RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF;
 339
 340         mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
 341                  sizeof(struct rte_pktmbuf_pool_private), socket_id, 0);
 342         if (mp == NULL)
 343                 return NULL;
 344
 345         mp_ops_name = rte_mbuf_best_mempool_ops();
 346         ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
 347         if (ret != 0) {
 348                 RTE_LOG(ERR, MBUF, "error setting mempool handler\n");
 349                 rte_mempool_free(mp);
 350                 rte_errno = -ret;
 351                 return NULL;
 352         }
 353         rte_pktmbuf_pool_init(mp, &mbp_priv);
 354
 355         ret = rte_mempool_populate_default(mp);
 356         if (ret < 0) {
 357                 rte_mempool_free(mp);
 358                 rte_errno = -ret;
 359                 return NULL;
 360         }
 361
 362         init_ctx = (struct rte_pktmbuf_extmem_init_ctx){
 363                 .ext_mem = ext_mem,
 364                 .ext_num = ext_num,
 365                 .ext = 0,
 366                 .off = 0,
 367         };
 368         rte_mempool_obj_iter(mp, __rte_pktmbuf_init_extmem, &init_ctx);
 369
 370         return mp;
 371 }
 372
 373 /* do some sanity checks on a mbuf: panic if it fails */
 374 void
 375 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header)
 376 {
 377         const char *reason;
 378
 379         if (rte_mbuf_check(m, is_header, &reason))
 380                 rte_panic("%s\n", reason);
 381 }
 382
 383 int rte_mbuf_check(const struct rte_mbuf *m, int is_header,
 384                    const char **reason)
 385 {
 386         unsigned int nb_segs, pkt_len;
 387
 388         if (m == NULL) {
 389                 *reason = "mbuf is NULL";
 390                 return -1;
 391         }
 392
 393         /* generic checks */
 394         if (m->pool == NULL) {
 395                 *reason = "bad mbuf pool";
 396                 return -1;
 397         }
 398         if (m->buf_iova == 0) {
 399                 *reason = "bad IO addr";
 400                 return -1;
 401         }
 402         if (m->buf_addr == NULL) {
 403                 *reason = "bad virt addr";
 404                 return -1;
 405         }
 406
 407         uint16_t cnt = rte_mbuf_refcnt_read(m);
 408         if ((cnt == 0) || (cnt == UINT16_MAX)) {
 409                 *reason = "bad ref cnt";
 410                 return -1;
 411         }
 412
 413         /* nothing to check for sub-segments */
 414         if (is_header == 0)
 415                 return 0;
 416
 417         /* data_len is supposed to be not more than pkt_len */
 418         if (m->data_len > m->pkt_len) {
 419                 *reason = "bad data_len";
 420                 return -1;
 421         }
 422
 423         nb_segs = m->nb_segs;
 424         pkt_len = m->pkt_len;
 425
 426         do {
 427                 if (m->data_off > m->buf_len) {
 428                         *reason = "data offset too big in mbuf segment";
 429                         return -1;
 430                 }
 431                 if (m->data_off + m->data_len > m->buf_len) {
 432                         *reason = "data length too big in mbuf segment";
 433                         return -1;
 434                 }
 435                 nb_segs -= 1;
 436                 pkt_len -= m->data_len;
 437         } while ((m = m->next) != NULL);
 438
 439         if (nb_segs) {
 440                 *reason = "bad nb_segs";
 441                 return -1;
 442         }
 443         if (pkt_len) {
 444                 *reason = "bad pkt_len";
 445                 return -1;
 446         }
 447
 448         return 0;
 449 }
 450
 451 /**
 452  * @internal helper function for freeing a bulk of packet mbuf segments
 453  * via an array holding the packet mbuf segments from the same mempool
 454  * pending to be freed.
 455  *
 456  * @param m
 457  *  The packet mbuf segment to be freed.
 458  * @param pending
 459  *  Pointer to the array of packet mbuf segments pending to be freed.
 460  * @param nb_pending
 461  *  Pointer to the number of elements held in the array.
 462  * @param pending_sz
 463  *  Number of elements the array can hold.
 464  *  Note: The compiler should optimize this parameter away when using a
 465  *  constant value, such as RTE_PKTMBUF_FREE_PENDING_SZ.
 466  */
 467 static void
 468 __rte_pktmbuf_free_seg_via_array(struct rte_mbuf *m,
 469         struct rte_mbuf ** const pending, unsigned int * const nb_pending,
 470         const unsigned int pending_sz)
 471 {
 472         m = rte_pktmbuf_prefree_seg(m);
 473         if (likely(m != NULL)) {
 474                 if (*nb_pending == pending_sz ||
 475                     (*nb_pending > 0 && m->pool != pending[0]->pool)) {
 476                         rte_mempool_put_bulk(pending[0]->pool,
 477                                         (void **)pending, *nb_pending);
 478                         *nb_pending = 0;
 479                 }
 480
 481                 pending[(*nb_pending)++] = m;
 482         }
 483 }
 484
 485 /**
 486  * Size of the array holding mbufs from the same mempool pending to be freed
 487  * in bulk.
 488  */
 489 #define RTE_PKTMBUF_FREE_PENDING_SZ 64
 490
 491 /* Free a bulk of packet mbufs back into their original mempools. */
 492 void rte_pktmbuf_free_bulk(struct rte_mbuf **mbufs, unsigned int count)
 493 {
 494         struct rte_mbuf *m, *m_next, *pending[RTE_PKTMBUF_FREE_PENDING_SZ];
 495         unsigned int idx, nb_pending = 0;
 496
 497         for (idx = 0; idx < count; idx++) {
 498                 m = mbufs[idx];
 499                 if (unlikely(m == NULL))
 500                         continue;
 501
 502                 __rte_mbuf_sanity_check(m, 1);
 503
 504                 do {
 505                         m_next = m->next;
 506                         __rte_pktmbuf_free_seg_via_array(m,
 507                                         pending, &nb_pending,
 508                                         RTE_PKTMBUF_FREE_PENDING_SZ);
 509                         m = m_next;
 510                 } while (m != NULL);
 511         }
 512
 513         if (nb_pending > 0)
 514                 rte_mempool_put_bulk(pending[0]->pool, (void **)pending, nb_pending);
 515 }
 516
 517 /* Creates a shallow copy of mbuf */
 518 struct rte_mbuf *
 519 rte_pktmbuf_clone(struct rte_mbuf *md, struct rte_mempool *mp)
 520 {
 521         struct rte_mbuf *mc, *mi, **prev;
 522         uint32_t pktlen;
 523         uint16_t nseg;
 524
 525         mc = rte_pktmbuf_alloc(mp);
 526         if (unlikely(mc == NULL))
 527                 return NULL;
 528
 529         mi = mc;
 530         prev = &mi->next;
 531         pktlen = md->pkt_len;
 532         nseg = 0;
 533
 534         do {
 535                 nseg++;
 536                 rte_pktmbuf_attach(mi, md);
 537                 *prev = mi;
 538                 prev = &mi->next;
 539         } while ((md = md->next) != NULL &&
 540             (mi = rte_pktmbuf_alloc(mp)) != NULL);
 541
 542         *prev = NULL;
 543         mc->nb_segs = nseg;
 544         mc->pkt_len = pktlen;
 545
 546         /* Allocation of new indirect segment failed */
 547         if (unlikely(mi == NULL)) {
 548                 rte_pktmbuf_free(mc);
 549                 return NULL;
 550         }
 551
 552         __rte_mbuf_sanity_check(mc, 1);
 553         return mc;
 554 }
 555
 556 /* convert multi-segment mbuf to single mbuf */
 557 int
 558 __rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
 559 {
 560         size_t seg_len, copy_len;
 561         struct rte_mbuf *m;
 562         struct rte_mbuf *m_next;
 563         char *buffer;
 564
 565         /* Extend first segment to the total packet length */
 566         copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
 567
 568         if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
 569                 return -1;
 570
 571         buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
 572         mbuf->data_len = (uint16_t)(mbuf->pkt_len);
 573
 574         /* Append data from next segments to the first one */
 575         m = mbuf->next;
 576         while (m != NULL) {
 577                 m_next = m->next;
 578
 579                 seg_len = rte_pktmbuf_data_len(m);
 580                 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
 581                 buffer += seg_len;
 582
 583                 rte_pktmbuf_free_seg(m);
 584                 m = m_next;
 585         }
 586
 587         mbuf->next = NULL;
 588         mbuf->nb_segs = 1;
 589
 590         return 0;
 591 }
 592
 593 /* Create a deep copy of mbuf */
 594 struct rte_mbuf *
 595 rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 596                  uint32_t off, uint32_t len)
 597 {
 598         const struct rte_mbuf *seg = m;
 599         struct rte_mbuf *mc, *m_last, **prev;
 600
 601         /* garbage in check */
 602         __rte_mbuf_sanity_check(m, 1);
 603
 604         /* check for request to copy at offset past end of mbuf */
 605         if (unlikely(off >= m->pkt_len))
 606                 return NULL;
 607
 608         mc = rte_pktmbuf_alloc(mp);
 609         if (unlikely(mc == NULL))
 610                 return NULL;
 611
 612         /* truncate requested length to available data */
 613         if (len > m->pkt_len - off)
 614                 len = m->pkt_len - off;
 615
 616         __rte_pktmbuf_copy_hdr(mc, m);
 617
 618         /* copied mbuf is not indirect or external */
 619         mc->ol_flags = m->ol_flags & ~(IND_ATTACHED_MBUF|EXT_ATTACHED_MBUF);
 620
 621         prev = &mc->next;
 622         m_last = mc;
 623         while (len > 0) {
 624                 uint32_t copy_len;
 625
 626                 /* skip leading mbuf segments */
 627                 while (off >= seg->data_len) {
 628                         off -= seg->data_len;
 629                         seg = seg->next;
 630                 }
 631
 632                 /* current buffer is full, chain a new one */
 633                 if (rte_pktmbuf_tailroom(m_last) == 0) {
 634                         m_last = rte_pktmbuf_alloc(mp);
 635                         if (unlikely(m_last == NULL)) {
 636                                 rte_pktmbuf_free(mc);
 637                                 return NULL;
 638                         }
 639                         ++mc->nb_segs;
 640                         *prev = m_last;
 641                         prev = &m_last->next;
 642                 }
 643
 644                 /*
 645                  * copy the min of data in input segment (seg)
 646                  * vs space available in output (m_last)
 647                  */
 648                 copy_len = RTE_MIN(seg->data_len - off, len);
 649                 if (copy_len > rte_pktmbuf_tailroom(m_last))
 650                         copy_len = rte_pktmbuf_tailroom(m_last);
 651
 652                 /* append from seg to m_last */
 653                 rte_memcpy(rte_pktmbuf_mtod_offset(m_last, char *,
 654                                                    m_last->data_len),
 655                            rte_pktmbuf_mtod_offset(seg, char *, off),
 656                            copy_len);
 657
 658                 /* update offsets and lengths */
 659                 m_last->data_len += copy_len;
 660                 mc->pkt_len += copy_len;
 661                 off += copy_len;
 662                 len -= copy_len;
 663         }
 664
 665         /* garbage out check */
 666         __rte_mbuf_sanity_check(mc, 1);
 667         return mc;
 668 }
 669
 670 /* dump a mbuf on console */
 671 void
 672 rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len)
 673 {
 674         unsigned int len;
 675         unsigned int nb_segs;
 676
 677         __rte_mbuf_sanity_check(m, 1);
 678
 679         fprintf(f, "dump mbuf at %p, iova=%#"PRIx64", buf_len=%u\n",
 680                 m, m->buf_iova, m->buf_len);
 681         fprintf(f, "  pkt_len=%u, ol_flags=%#"PRIx64", nb_segs=%u, port=%u",
 682                 m->pkt_len, m->ol_flags, m->nb_segs, m->port);
 683
 684         if (m->ol_flags & (PKT_RX_VLAN | PKT_TX_VLAN))
 685                 fprintf(f, ", vlan_tci=%u", m->vlan_tci);
 686
 687         fprintf(f, ", ptype=%#"PRIx32"\n", m->packet_type);
 688
 689         nb_segs = m->nb_segs;
 690
 691         while (m && nb_segs != 0) {
 692                 __rte_mbuf_sanity_check(m, 0);
 693
 694                 fprintf(f, "  segment at %p, data=%p, len=%u, off=%u, refcnt=%u\n",
 695                         m, rte_pktmbuf_mtod(m, void *),
 696                         m->data_len, m->data_off, rte_mbuf_refcnt_read(m));
 697
 698                 len = dump_len;
 699                 if (len > m->data_len)
 700                         len = m->data_len;
 701                 if (len != 0)
 702                         rte_hexdump(f, NULL, rte_pktmbuf_mtod(m, void *), len);
 703                 dump_len -= len;
 704                 m = m->next;
 705                 nb_segs --;
 706         }
 707 }
 708
 709 /* read len data bytes in a mbuf at specified offset (internal) */
 710 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
 711         uint32_t len, void *buf)
 712 {
 713         const struct rte_mbuf *seg = m;
 714         uint32_t buf_off = 0, copy_len;
 715
 716         if (off + len > rte_pktmbuf_pkt_len(m))
 717                 return NULL;
 718
 719         while (off >= rte_pktmbuf_data_len(seg)) {
 720                 off -= rte_pktmbuf_data_len(seg);
 721                 seg = seg->next;
 722         }
 723
 724         if (off + len <= rte_pktmbuf_data_len(seg))
 725                 return rte_pktmbuf_mtod_offset(seg, char *, off);
 726
 727         /* rare case: header is split among several segments */
 728         while (len > 0) {
 729                 copy_len = rte_pktmbuf_data_len(seg) - off;
 730                 if (copy_len > len)
 731                         copy_len = len;
 732                 rte_memcpy((char *)buf + buf_off,
 733                         rte_pktmbuf_mtod_offset(seg, char *, off), copy_len);
 734                 off = 0;
 735                 buf_off += copy_len;
 736                 len -= copy_len;
 737                 seg = seg->next;
 738         }
 739
 740         return buf;
 741 }
 742
 743 /*
 744  * Get the name of a RX offload flag. Must be kept synchronized with flag
 745  * definitions in rte_mbuf.h.
 746  */
 747 const char *rte_get_rx_ol_flag_name(uint64_t mask)
 748 {
 749         switch (mask) {
 750         case PKT_RX_VLAN: return "PKT_RX_VLAN";
 751         case PKT_RX_RSS_HASH: return "PKT_RX_RSS_HASH";
 752         case PKT_RX_FDIR: return "PKT_RX_FDIR";
 753         case PKT_RX_L4_CKSUM_BAD: return "PKT_RX_L4_CKSUM_BAD";
 754         case PKT_RX_L4_CKSUM_GOOD: return "PKT_RX_L4_CKSUM_GOOD";
 755         case PKT_RX_L4_CKSUM_NONE: return "PKT_RX_L4_CKSUM_NONE";
 756         case PKT_RX_IP_CKSUM_BAD: return "PKT_RX_IP_CKSUM_BAD";
 757         case PKT_RX_IP_CKSUM_GOOD: return "PKT_RX_IP_CKSUM_GOOD";
 758         case PKT_RX_IP_CKSUM_NONE: return "PKT_RX_IP_CKSUM_NONE";
 759         case PKT_RX_EIP_CKSUM_BAD: return "PKT_RX_EIP_CKSUM_BAD";
 760         case PKT_RX_VLAN_STRIPPED: return "PKT_RX_VLAN_STRIPPED";
 761         case PKT_RX_IEEE1588_PTP: return "PKT_RX_IEEE1588_PTP";
 762         case PKT_RX_IEEE1588_TMST: return "PKT_RX_IEEE1588_TMST";
 763         case PKT_RX_FDIR_ID: return "PKT_RX_FDIR_ID";
 764         case PKT_RX_FDIR_FLX: return "PKT_RX_FDIR_FLX";
 765         case PKT_RX_QINQ_STRIPPED: return "PKT_RX_QINQ_STRIPPED";
 766         case PKT_RX_QINQ: return "PKT_RX_QINQ";
 767         case PKT_RX_LRO: return "PKT_RX_LRO";
 768         case PKT_RX_TIMESTAMP: return "PKT_RX_TIMESTAMP";
 769         case PKT_RX_SEC_OFFLOAD: return "PKT_RX_SEC_OFFLOAD";
 770         case PKT_RX_SEC_OFFLOAD_FAILED: return "PKT_RX_SEC_OFFLOAD_FAILED";
 771         case PKT_RX_OUTER_L4_CKSUM_BAD: return "PKT_RX_OUTER_L4_CKSUM_BAD";
 772         case PKT_RX_OUTER_L4_CKSUM_GOOD: return "PKT_RX_OUTER_L4_CKSUM_GOOD";
 773         case PKT_RX_OUTER_L4_CKSUM_INVALID:
 774                 return "PKT_RX_OUTER_L4_CKSUM_INVALID";
 775
 776         default: return NULL;
 777         }
 778 }
 779
 780 struct flag_mask {
 781         uint64_t flag;
 782         uint64_t mask;
 783         const char *default_name;
 784 };
 785
 786 /* write the list of rx ol flags in buffer buf */
 787 int
 788 rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen)
 789 {
 790         const struct flag_mask rx_flags[] = {
 791                 { PKT_RX_VLAN, PKT_RX_VLAN, NULL },
 792                 { PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, NULL },
 793                 { PKT_RX_FDIR, PKT_RX_FDIR, NULL },
 794                 { PKT_RX_L4_CKSUM_BAD, PKT_RX_L4_CKSUM_MASK, NULL },
 795                 { PKT_RX_L4_CKSUM_GOOD, PKT_RX_L4_CKSUM_MASK, NULL },
 796                 { PKT_RX_L4_CKSUM_NONE, PKT_RX_L4_CKSUM_MASK, NULL },
 797                 { PKT_RX_L4_CKSUM_UNKNOWN, PKT_RX_L4_CKSUM_MASK,
 798                   "PKT_RX_L4_CKSUM_UNKNOWN" },
 799                 { PKT_RX_IP_CKSUM_BAD, PKT_RX_IP_CKSUM_MASK, NULL },
 800                 { PKT_RX_IP_CKSUM_GOOD, PKT_RX_IP_CKSUM_MASK, NULL },
 801                 { PKT_RX_IP_CKSUM_NONE, PKT_RX_IP_CKSUM_MASK, NULL },
 802                 { PKT_RX_IP_CKSUM_UNKNOWN, PKT_RX_IP_CKSUM_MASK,
 803                   "PKT_RX_IP_CKSUM_UNKNOWN" },
 804                 { PKT_RX_EIP_CKSUM_BAD, PKT_RX_EIP_CKSUM_BAD, NULL },
 805                 { PKT_RX_VLAN_STRIPPED, PKT_RX_VLAN_STRIPPED, NULL },
 806                 { PKT_RX_IEEE1588_PTP, PKT_RX_IEEE1588_PTP, NULL },
 807                 { PKT_RX_IEEE1588_TMST, PKT_RX_IEEE1588_TMST, NULL },
 808                 { PKT_RX_FDIR_ID, PKT_RX_FDIR_ID, NULL },
 809                 { PKT_RX_FDIR_FLX, PKT_RX_FDIR_FLX, NULL },
 810                 { PKT_RX_QINQ_STRIPPED, PKT_RX_QINQ_STRIPPED, NULL },
 811                 { PKT_RX_LRO, PKT_RX_LRO, NULL },
 812                 { PKT_RX_TIMESTAMP, PKT_RX_TIMESTAMP, NULL },
 813                 { PKT_RX_SEC_OFFLOAD, PKT_RX_SEC_OFFLOAD, NULL },
 814                 { PKT_RX_SEC_OFFLOAD_FAILED, PKT_RX_SEC_OFFLOAD_FAILED, NULL },
 815                 { PKT_RX_QINQ, PKT_RX_QINQ, NULL },
 816                 { PKT_RX_OUTER_L4_CKSUM_BAD, PKT_RX_OUTER_L4_CKSUM_MASK, NULL },
 817                 { PKT_RX_OUTER_L4_CKSUM_GOOD, PKT_RX_OUTER_L4_CKSUM_MASK,
 818                   NULL },
 819                 { PKT_RX_OUTER_L4_CKSUM_INVALID, PKT_RX_OUTER_L4_CKSUM_MASK,
 820                   NULL },
 821                 { PKT_RX_OUTER_L4_CKSUM_UNKNOWN, PKT_RX_OUTER_L4_CKSUM_MASK,
 822                   "PKT_RX_OUTER_L4_CKSUM_UNKNOWN" },
 823         };
 824         const char *name;
 825         unsigned int i;
 826         int ret;
 827
 828         if (buflen == 0)
 829                 return -1;
 830
 831         buf[0] = '\0';
 832         for (i = 0; i < RTE_DIM(rx_flags); i++) {
 833                 if ((mask & rx_flags[i].mask) != rx_flags[i].flag)
 834                         continue;
 835                 name = rte_get_rx_ol_flag_name(rx_flags[i].flag);
 836                 if (name == NULL)
 837                         name = rx_flags[i].default_name;
 838                 ret = snprintf(buf, buflen, "%s ", name);
 839                 if (ret < 0)
 840                         return -1;
 841                 if ((size_t)ret >= buflen)
 842                         return -1;
 843                 buf += ret;
 844                 buflen -= ret;
 845         }
 846
 847         return 0;
 848 }
 849
 850 /*
 851  * Get the name of a TX offload flag. Must be kept synchronized with flag
 852  * definitions in rte_mbuf.h.
 853  */
 854 const char *rte_get_tx_ol_flag_name(uint64_t mask)
 855 {
 856         switch (mask) {
 857         case PKT_TX_VLAN: return "PKT_TX_VLAN";
 858         case PKT_TX_IP_CKSUM: return "PKT_TX_IP_CKSUM";
 859         case PKT_TX_TCP_CKSUM: return "PKT_TX_TCP_CKSUM";
 860         case PKT_TX_SCTP_CKSUM: return "PKT_TX_SCTP_CKSUM";
 861         case PKT_TX_UDP_CKSUM: return "PKT_TX_UDP_CKSUM";
 862         case PKT_TX_IEEE1588_TMST: return "PKT_TX_IEEE1588_TMST";
 863         case PKT_TX_TCP_SEG: return "PKT_TX_TCP_SEG";
 864         case PKT_TX_IPV4: return "PKT_TX_IPV4";
 865         case PKT_TX_IPV6: return "PKT_TX_IPV6";
 866         case PKT_TX_OUTER_IP_CKSUM: return "PKT_TX_OUTER_IP_CKSUM";
 867         case PKT_TX_OUTER_IPV4: return "PKT_TX_OUTER_IPV4";
 868         case PKT_TX_OUTER_IPV6: return "PKT_TX_OUTER_IPV6";
 869         case PKT_TX_TUNNEL_VXLAN: return "PKT_TX_TUNNEL_VXLAN";
 870         case PKT_TX_TUNNEL_GTP: return "PKT_TX_TUNNEL_GTP";
 871         case PKT_TX_TUNNEL_GRE: return "PKT_TX_TUNNEL_GRE";
 872         case PKT_TX_TUNNEL_IPIP: return "PKT_TX_TUNNEL_IPIP";
 873         case PKT_TX_TUNNEL_GENEVE: return "PKT_TX_TUNNEL_GENEVE";
 874         case PKT_TX_TUNNEL_MPLSINUDP: return "PKT_TX_TUNNEL_MPLSINUDP";
 875         case PKT_TX_TUNNEL_VXLAN_GPE: return "PKT_TX_TUNNEL_VXLAN_GPE";
 876         case PKT_TX_TUNNEL_IP: return "PKT_TX_TUNNEL_IP";
 877         case PKT_TX_TUNNEL_UDP: return "PKT_TX_TUNNEL_UDP";
 878         case PKT_TX_QINQ: return "PKT_TX_QINQ";
 879         case PKT_TX_MACSEC: return "PKT_TX_MACSEC";
 880         case PKT_TX_SEC_OFFLOAD: return "PKT_TX_SEC_OFFLOAD";
 881         case PKT_TX_UDP_SEG: return "PKT_TX_UDP_SEG";
 882         case PKT_TX_OUTER_UDP_CKSUM: return "PKT_TX_OUTER_UDP_CKSUM";
 883         default: return NULL;
 884         }
 885 }
 886
 887 /* write the list of tx ol flags in buffer buf */
 888 int
 889 rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen)
 890 {
 891         const struct flag_mask tx_flags[] = {
 892                 { PKT_TX_VLAN, PKT_TX_VLAN, NULL },
 893                 { PKT_TX_IP_CKSUM, PKT_TX_IP_CKSUM, NULL },
 894                 { PKT_TX_TCP_CKSUM, PKT_TX_L4_MASK, NULL },
 895                 { PKT_TX_SCTP_CKSUM, PKT_TX_L4_MASK, NULL },
 896                 { PKT_TX_UDP_CKSUM, PKT_TX_L4_MASK, NULL },
 897                 { PKT_TX_L4_NO_CKSUM, PKT_TX_L4_MASK, "PKT_TX_L4_NO_CKSUM" },
 898                 { PKT_TX_IEEE1588_TMST, PKT_TX_IEEE1588_TMST, NULL },
 899                 { PKT_TX_TCP_SEG, PKT_TX_TCP_SEG, NULL },
 900                 { PKT_TX_IPV4, PKT_TX_IPV4, NULL },
 901                 { PKT_TX_IPV6, PKT_TX_IPV6, NULL },
 902                 { PKT_TX_OUTER_IP_CKSUM, PKT_TX_OUTER_IP_CKSUM, NULL },
 903                 { PKT_TX_OUTER_IPV4, PKT_TX_OUTER_IPV4, NULL },
 904                 { PKT_TX_OUTER_IPV6, PKT_TX_OUTER_IPV6, NULL },
 905                 { PKT_TX_TUNNEL_VXLAN, PKT_TX_TUNNEL_MASK, NULL },
 906                 { PKT_TX_TUNNEL_GTP, PKT_TX_TUNNEL_MASK, NULL },
 907                 { PKT_TX_TUNNEL_GRE, PKT_TX_TUNNEL_MASK, NULL },
 908                 { PKT_TX_TUNNEL_IPIP, PKT_TX_TUNNEL_MASK, NULL },
 909                 { PKT_TX_TUNNEL_GENEVE, PKT_TX_TUNNEL_MASK, NULL },
 910                 { PKT_TX_TUNNEL_MPLSINUDP, PKT_TX_TUNNEL_MASK, NULL },
 911                 { PKT_TX_TUNNEL_VXLAN_GPE, PKT_TX_TUNNEL_MASK, NULL },
 912                 { PKT_TX_TUNNEL_IP, PKT_TX_TUNNEL_MASK, NULL },
 913                 { PKT_TX_TUNNEL_UDP, PKT_TX_TUNNEL_MASK, NULL },
 914                 { PKT_TX_QINQ, PKT_TX_QINQ, NULL },
 915                 { PKT_TX_MACSEC, PKT_TX_MACSEC, NULL },
 916                 { PKT_TX_SEC_OFFLOAD, PKT_TX_SEC_OFFLOAD, NULL },
 917                 { PKT_TX_UDP_SEG, PKT_TX_UDP_SEG, NULL },
 918                 { PKT_TX_OUTER_UDP_CKSUM, PKT_TX_OUTER_UDP_CKSUM, NULL },
 919         };
 920         const char *name;
 921         unsigned int i;
 922         int ret;
 923
 924         if (buflen == 0)
 925                 return -1;
 926
 927         buf[0] = '\0';
 928         for (i = 0; i < RTE_DIM(tx_flags); i++) {
 929                 if ((mask & tx_flags[i].mask) != tx_flags[i].flag)
 930                         continue;
 931                 name = rte_get_tx_ol_flag_name(tx_flags[i].flag);
 932                 if (name == NULL)
 933                         name = tx_flags[i].default_name;
 934                 ret = snprintf(buf, buflen, "%s ", name);
 935                 if (ret < 0)
 936                         return -1;
 937                 if ((size_t)ret >= buflen)
 938                         return -1;
 939                 buf += ret;
 940                 buflen -= ret;
 941         }
 942
 943         return 0;
 944 }