linux/fs/nfs/objlayout/objlayout.c
<<
>>
Prefs
   1/*
   2 *  pNFS Objects layout driver high level definitions
   3 *
   4 *  Copyright (C) 2007 Panasas Inc. [year of first publication]
   5 *  All rights reserved.
   6 *
   7 *  Benny Halevy <bhalevy@panasas.com>
   8 *  Boaz Harrosh <bharrosh@panasas.com>
   9 *
  10 *  This program is free software; you can redistribute it and/or modify
  11 *  it under the terms of the GNU General Public License version 2
  12 *  See the file COPYING included with this distribution for more details.
  13 *
  14 *  Redistribution and use in source and binary forms, with or without
  15 *  modification, are permitted provided that the following conditions
  16 *  are met:
  17 *
  18 *  1. Redistributions of source code must retain the above copyright
  19 *     notice, this list of conditions and the following disclaimer.
  20 *  2. Redistributions in binary form must reproduce the above copyright
  21 *     notice, this list of conditions and the following disclaimer in the
  22 *     documentation and/or other materials provided with the distribution.
  23 *  3. Neither the name of the Panasas company nor the names of its
  24 *     contributors may be used to endorse or promote products derived
  25 *     from this software without specific prior written permission.
  26 *
  27 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  28 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  29 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  30 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  31 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  32 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  33 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  34 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  35 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  36 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  37 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  38 */
  39
  40#include <linux/kmod.h>
  41#include <linux/moduleparam.h>
  42#include <linux/ratelimit.h>
  43#include <scsi/osd_initiator.h>
  44#include "objlayout.h"
  45
  46#define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  47/*
  48 * Create a objlayout layout structure for the given inode and return it.
  49 */
  50struct pnfs_layout_hdr *
  51objlayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  52{
  53        struct objlayout *objlay;
  54
  55        objlay = kzalloc(sizeof(struct objlayout), gfp_flags);
  56        if (objlay) {
  57                spin_lock_init(&objlay->lock);
  58                INIT_LIST_HEAD(&objlay->err_list);
  59        }
  60        dprintk("%s: Return %p\n", __func__, objlay);
  61        return &objlay->pnfs_layout;
  62}
  63
  64/*
  65 * Free an objlayout layout structure
  66 */
  67void
  68objlayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  69{
  70        struct objlayout *objlay = OBJLAYOUT(lo);
  71
  72        dprintk("%s: objlay %p\n", __func__, objlay);
  73
  74        WARN_ON(!list_empty(&objlay->err_list));
  75        kfree(objlay);
  76}
  77
  78/*
  79 * Unmarshall layout and store it in pnfslay.
  80 */
  81struct pnfs_layout_segment *
  82objlayout_alloc_lseg(struct pnfs_layout_hdr *pnfslay,
  83                     struct nfs4_layoutget_res *lgr,
  84                     gfp_t gfp_flags)
  85{
  86        int status = -ENOMEM;
  87        struct xdr_stream stream;
  88        struct xdr_buf buf = {
  89                .pages =  lgr->layoutp->pages,
  90                .page_len =  lgr->layoutp->len,
  91                .buflen =  lgr->layoutp->len,
  92                .len = lgr->layoutp->len,
  93        };
  94        struct page *scratch;
  95        struct pnfs_layout_segment *lseg;
  96
  97        dprintk("%s: Begin pnfslay %p\n", __func__, pnfslay);
  98
  99        scratch = alloc_page(gfp_flags);
 100        if (!scratch)
 101                goto err_nofree;
 102
 103        xdr_init_decode(&stream, &buf, NULL);
 104        xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 105
 106        status = objio_alloc_lseg(&lseg, pnfslay, &lgr->range, &stream, gfp_flags);
 107        if (unlikely(status)) {
 108                dprintk("%s: objio_alloc_lseg Return err %d\n", __func__,
 109                        status);
 110                goto err;
 111        }
 112
 113        __free_page(scratch);
 114
 115        dprintk("%s: Return %p\n", __func__, lseg);
 116        return lseg;
 117
 118err:
 119        __free_page(scratch);
 120err_nofree:
 121        dprintk("%s: Err Return=>%d\n", __func__, status);
 122        return ERR_PTR(status);
 123}
 124
 125/*
 126 * Free a layout segement
 127 */
 128void
 129objlayout_free_lseg(struct pnfs_layout_segment *lseg)
 130{
 131        dprintk("%s: freeing layout segment %p\n", __func__, lseg);
 132
 133        if (unlikely(!lseg))
 134                return;
 135
 136        objio_free_lseg(lseg);
 137}
 138
 139/*
 140 * I/O Operations
 141 */
 142static inline u64
 143end_offset(u64 start, u64 len)
 144{
 145        u64 end;
 146
 147        end = start + len;
 148        return end >= start ? end : NFS4_MAX_UINT64;
 149}
 150
 151static void _fix_verify_io_params(struct pnfs_layout_segment *lseg,
 152                           struct page ***p_pages, unsigned *p_pgbase,
 153                           u64 offset, unsigned long count)
 154{
 155        u64 lseg_end_offset;
 156
 157        BUG_ON(offset < lseg->pls_range.offset);
 158        lseg_end_offset = end_offset(lseg->pls_range.offset,
 159                                     lseg->pls_range.length);
 160        BUG_ON(offset >= lseg_end_offset);
 161        WARN_ON(offset + count > lseg_end_offset);
 162
 163        if (*p_pgbase > PAGE_SIZE) {
 164                dprintk("%s: pgbase(0x%x) > PAGE_SIZE\n", __func__, *p_pgbase);
 165                *p_pages += *p_pgbase >> PAGE_SHIFT;
 166                *p_pgbase &= ~PAGE_MASK;
 167        }
 168}
 169
 170/*
 171 * I/O done common code
 172 */
 173static void
 174objlayout_iodone(struct objlayout_io_res *oir)
 175{
 176        if (likely(oir->status >= 0)) {
 177                objio_free_result(oir);
 178        } else {
 179                struct objlayout *objlay = oir->objlay;
 180
 181                spin_lock(&objlay->lock);
 182                objlay->delta_space_valid = OBJ_DSU_INVALID;
 183                list_add(&objlay->err_list, &oir->err_list);
 184                spin_unlock(&objlay->lock);
 185        }
 186}
 187
 188/*
 189 * objlayout_io_set_result - Set an osd_error code on a specific osd comp.
 190 *
 191 * The @index component IO failed (error returned from target). Register
 192 * the error for later reporting at layout-return.
 193 */
 194void
 195objlayout_io_set_result(struct objlayout_io_res *oir, unsigned index,
 196                        struct pnfs_osd_objid *pooid, int osd_error,
 197                        u64 offset, u64 length, bool is_write)
 198{
 199        struct pnfs_osd_ioerr *ioerr = &oir->ioerrs[index];
 200
 201        BUG_ON(index >= oir->num_comps);
 202        if (osd_error) {
 203                ioerr->oer_component = *pooid;
 204                ioerr->oer_comp_offset = offset;
 205                ioerr->oer_comp_length = length;
 206                ioerr->oer_iswrite = is_write;
 207                ioerr->oer_errno = osd_error;
 208
 209                dprintk("%s: err[%d]: errno=%d is_write=%d dev(%llx:%llx) "
 210                        "par=0x%llx obj=0x%llx offset=0x%llx length=0x%llx\n",
 211                        __func__, index, ioerr->oer_errno,
 212                        ioerr->oer_iswrite,
 213                        _DEVID_LO(&ioerr->oer_component.oid_device_id),
 214                        _DEVID_HI(&ioerr->oer_component.oid_device_id),
 215                        ioerr->oer_component.oid_partition_id,
 216                        ioerr->oer_component.oid_object_id,
 217                        ioerr->oer_comp_offset,
 218                        ioerr->oer_comp_length);
 219        } else {
 220                /* User need not call if no error is reported */
 221                ioerr->oer_errno = 0;
 222        }
 223}
 224
 225/* Function scheduled on rpc workqueue to call ->nfs_readlist_complete().
 226 * This is because the osd completion is called with ints-off from
 227 * the block layer
 228 */
 229static void _rpc_read_complete(struct work_struct *work)
 230{
 231        struct rpc_task *task;
 232        struct nfs_read_data *rdata;
 233
 234        dprintk("%s enter\n", __func__);
 235        task = container_of(work, struct rpc_task, u.tk_work);
 236        rdata = container_of(task, struct nfs_read_data, task);
 237
 238        pnfs_ld_read_done(rdata);
 239}
 240
 241void
 242objlayout_read_done(struct objlayout_io_res *oir, ssize_t status, bool sync)
 243{
 244        struct nfs_read_data *rdata = oir->rpcdata;
 245
 246        oir->status = rdata->task.tk_status = status;
 247        if (status >= 0)
 248                rdata->res.count = status;
 249        else
 250                rdata->header->pnfs_error = status;
 251        objlayout_iodone(oir);
 252        /* must not use oir after this point */
 253
 254        dprintk("%s: Return status=%zd eof=%d sync=%d\n", __func__,
 255                status, rdata->res.eof, sync);
 256
 257        if (sync)
 258                pnfs_ld_read_done(rdata);
 259        else {
 260                INIT_WORK(&rdata->task.u.tk_work, _rpc_read_complete);
 261                schedule_work(&rdata->task.u.tk_work);
 262        }
 263}
 264
 265/*
 266 * Perform sync or async reads.
 267 */
 268enum pnfs_try_status
 269objlayout_read_pagelist(struct nfs_read_data *rdata)
 270{
 271        struct nfs_pgio_header *hdr = rdata->header;
 272        struct inode *inode = hdr->inode;
 273        loff_t offset = rdata->args.offset;
 274        size_t count = rdata->args.count;
 275        int err;
 276        loff_t eof;
 277
 278        eof = i_size_read(inode);
 279        if (unlikely(offset + count > eof)) {
 280                if (offset >= eof) {
 281                        err = 0;
 282                        rdata->res.count = 0;
 283                        rdata->res.eof = 1;
 284                        /*FIXME: do we need to call pnfs_ld_read_done() */
 285                        goto out;
 286                }
 287                count = eof - offset;
 288        }
 289
 290        rdata->res.eof = (offset + count) >= eof;
 291        _fix_verify_io_params(hdr->lseg, &rdata->args.pages,
 292                              &rdata->args.pgbase,
 293                              rdata->args.offset, rdata->args.count);
 294
 295        dprintk("%s: inode(%lx) offset 0x%llx count 0x%Zx eof=%d\n",
 296                __func__, inode->i_ino, offset, count, rdata->res.eof);
 297
 298        err = objio_read_pagelist(rdata);
 299 out:
 300        if (unlikely(err)) {
 301                hdr->pnfs_error = err;
 302                dprintk("%s: Returned Error %d\n", __func__, err);
 303                return PNFS_NOT_ATTEMPTED;
 304        }
 305        return PNFS_ATTEMPTED;
 306}
 307
 308/* Function scheduled on rpc workqueue to call ->nfs_writelist_complete().
 309 * This is because the osd completion is called with ints-off from
 310 * the block layer
 311 */
 312static void _rpc_write_complete(struct work_struct *work)
 313{
 314        struct rpc_task *task;
 315        struct nfs_write_data *wdata;
 316
 317        dprintk("%s enter\n", __func__);
 318        task = container_of(work, struct rpc_task, u.tk_work);
 319        wdata = container_of(task, struct nfs_write_data, task);
 320
 321        pnfs_ld_write_done(wdata);
 322}
 323
 324void
 325objlayout_write_done(struct objlayout_io_res *oir, ssize_t status, bool sync)
 326{
 327        struct nfs_write_data *wdata = oir->rpcdata;
 328
 329        oir->status = wdata->task.tk_status = status;
 330        if (status >= 0) {
 331                wdata->res.count = status;
 332                wdata->verf.committed = oir->committed;
 333        } else {
 334                wdata->header->pnfs_error = status;
 335        }
 336        objlayout_iodone(oir);
 337        /* must not use oir after this point */
 338
 339        dprintk("%s: Return status %zd committed %d sync=%d\n", __func__,
 340                status, wdata->verf.committed, sync);
 341
 342        if (sync)
 343                pnfs_ld_write_done(wdata);
 344        else {
 345                INIT_WORK(&wdata->task.u.tk_work, _rpc_write_complete);
 346                schedule_work(&wdata->task.u.tk_work);
 347        }
 348}
 349
 350/*
 351 * Perform sync or async writes.
 352 */
 353enum pnfs_try_status
 354objlayout_write_pagelist(struct nfs_write_data *wdata,
 355                         int how)
 356{
 357        struct nfs_pgio_header *hdr = wdata->header;
 358        int err;
 359
 360        _fix_verify_io_params(hdr->lseg, &wdata->args.pages,
 361                              &wdata->args.pgbase,
 362                              wdata->args.offset, wdata->args.count);
 363
 364        err = objio_write_pagelist(wdata, how);
 365        if (unlikely(err)) {
 366                hdr->pnfs_error = err;
 367                dprintk("%s: Returned Error %d\n", __func__, err);
 368                return PNFS_NOT_ATTEMPTED;
 369        }
 370        return PNFS_ATTEMPTED;
 371}
 372
 373void
 374objlayout_encode_layoutcommit(struct pnfs_layout_hdr *pnfslay,
 375                              struct xdr_stream *xdr,
 376                              const struct nfs4_layoutcommit_args *args)
 377{
 378        struct objlayout *objlay = OBJLAYOUT(pnfslay);
 379        struct pnfs_osd_layoutupdate lou;
 380        __be32 *start;
 381
 382        dprintk("%s: Begin\n", __func__);
 383
 384        spin_lock(&objlay->lock);
 385        lou.dsu_valid = (objlay->delta_space_valid == OBJ_DSU_VALID);
 386        lou.dsu_delta = objlay->delta_space_used;
 387        objlay->delta_space_used = 0;
 388        objlay->delta_space_valid = OBJ_DSU_INIT;
 389        lou.olu_ioerr_flag = !list_empty(&objlay->err_list);
 390        spin_unlock(&objlay->lock);
 391
 392        start = xdr_reserve_space(xdr, 4);
 393
 394        BUG_ON(pnfs_osd_xdr_encode_layoutupdate(xdr, &lou));
 395
 396        *start = cpu_to_be32((xdr->p - start - 1) * 4);
 397
 398        dprintk("%s: Return delta_space_used %lld err %d\n", __func__,
 399                lou.dsu_delta, lou.olu_ioerr_flag);
 400}
 401
 402static int
 403err_prio(u32 oer_errno)
 404{
 405        switch (oer_errno) {
 406        case 0:
 407                return 0;
 408
 409        case PNFS_OSD_ERR_RESOURCE:
 410                return OSD_ERR_PRI_RESOURCE;
 411        case PNFS_OSD_ERR_BAD_CRED:
 412                return OSD_ERR_PRI_BAD_CRED;
 413        case PNFS_OSD_ERR_NO_ACCESS:
 414                return OSD_ERR_PRI_NO_ACCESS;
 415        case PNFS_OSD_ERR_UNREACHABLE:
 416                return OSD_ERR_PRI_UNREACHABLE;
 417        case PNFS_OSD_ERR_NOT_FOUND:
 418                return OSD_ERR_PRI_NOT_FOUND;
 419        case PNFS_OSD_ERR_NO_SPACE:
 420                return OSD_ERR_PRI_NO_SPACE;
 421        default:
 422                WARN_ON(1);
 423                /* fallthrough */
 424        case PNFS_OSD_ERR_EIO:
 425                return OSD_ERR_PRI_EIO;
 426        }
 427}
 428
 429static void
 430merge_ioerr(struct pnfs_osd_ioerr *dest_err,
 431            const struct pnfs_osd_ioerr *src_err)
 432{
 433        u64 dest_end, src_end;
 434
 435        if (!dest_err->oer_errno) {
 436                *dest_err = *src_err;
 437                /* accumulated device must be blank */
 438                memset(&dest_err->oer_component.oid_device_id, 0,
 439                        sizeof(dest_err->oer_component.oid_device_id));
 440
 441                return;
 442        }
 443
 444        if (dest_err->oer_component.oid_partition_id !=
 445                                src_err->oer_component.oid_partition_id)
 446                dest_err->oer_component.oid_partition_id = 0;
 447
 448        if (dest_err->oer_component.oid_object_id !=
 449                                src_err->oer_component.oid_object_id)
 450                dest_err->oer_component.oid_object_id = 0;
 451
 452        if (dest_err->oer_comp_offset > src_err->oer_comp_offset)
 453                dest_err->oer_comp_offset = src_err->oer_comp_offset;
 454
 455        dest_end = end_offset(dest_err->oer_comp_offset,
 456                              dest_err->oer_comp_length);
 457        src_end =  end_offset(src_err->oer_comp_offset,
 458                              src_err->oer_comp_length);
 459        if (dest_end < src_end)
 460                dest_end = src_end;
 461
 462        dest_err->oer_comp_length = dest_end - dest_err->oer_comp_offset;
 463
 464        if ((src_err->oer_iswrite == dest_err->oer_iswrite) &&
 465            (err_prio(src_err->oer_errno) > err_prio(dest_err->oer_errno))) {
 466                        dest_err->oer_errno = src_err->oer_errno;
 467        } else if (src_err->oer_iswrite) {
 468                dest_err->oer_iswrite = true;
 469                dest_err->oer_errno = src_err->oer_errno;
 470        }
 471}
 472
 473static void
 474encode_accumulated_error(struct objlayout *objlay, __be32 *p)
 475{
 476        struct objlayout_io_res *oir, *tmp;
 477        struct pnfs_osd_ioerr accumulated_err = {.oer_errno = 0};
 478
 479        list_for_each_entry_safe(oir, tmp, &objlay->err_list, err_list) {
 480                unsigned i;
 481
 482                for (i = 0; i < oir->num_comps; i++) {
 483                        struct pnfs_osd_ioerr *ioerr = &oir->ioerrs[i];
 484
 485                        if (!ioerr->oer_errno)
 486                                continue;
 487
 488                        printk(KERN_ERR "NFS: %s: err[%d]: errno=%d "
 489                                "is_write=%d dev(%llx:%llx) par=0x%llx "
 490                                "obj=0x%llx offset=0x%llx length=0x%llx\n",
 491                                __func__, i, ioerr->oer_errno,
 492                                ioerr->oer_iswrite,
 493                                _DEVID_LO(&ioerr->oer_component.oid_device_id),
 494                                _DEVID_HI(&ioerr->oer_component.oid_device_id),
 495                                ioerr->oer_component.oid_partition_id,
 496                                ioerr->oer_component.oid_object_id,
 497                                ioerr->oer_comp_offset,
 498                                ioerr->oer_comp_length);
 499
 500                        merge_ioerr(&accumulated_err, ioerr);
 501                }
 502                list_del(&oir->err_list);
 503                objio_free_result(oir);
 504        }
 505
 506        pnfs_osd_xdr_encode_ioerr(p, &accumulated_err);
 507}
 508
 509void
 510objlayout_encode_layoutreturn(struct pnfs_layout_hdr *pnfslay,
 511                              struct xdr_stream *xdr,
 512                              const struct nfs4_layoutreturn_args *args)
 513{
 514        struct objlayout *objlay = OBJLAYOUT(pnfslay);
 515        struct objlayout_io_res *oir, *tmp;
 516        __be32 *start;
 517
 518        dprintk("%s: Begin\n", __func__);
 519        start = xdr_reserve_space(xdr, 4);
 520        BUG_ON(!start);
 521
 522        spin_lock(&objlay->lock);
 523
 524        list_for_each_entry_safe(oir, tmp, &objlay->err_list, err_list) {
 525                __be32 *last_xdr = NULL, *p;
 526                unsigned i;
 527                int res = 0;
 528
 529                for (i = 0; i < oir->num_comps; i++) {
 530                        struct pnfs_osd_ioerr *ioerr = &oir->ioerrs[i];
 531
 532                        if (!ioerr->oer_errno)
 533                                continue;
 534
 535                        dprintk("%s: err[%d]: errno=%d is_write=%d "
 536                                "dev(%llx:%llx) par=0x%llx obj=0x%llx "
 537                                "offset=0x%llx length=0x%llx\n",
 538                                __func__, i, ioerr->oer_errno,
 539                                ioerr->oer_iswrite,
 540                                _DEVID_LO(&ioerr->oer_component.oid_device_id),
 541                                _DEVID_HI(&ioerr->oer_component.oid_device_id),
 542                                ioerr->oer_component.oid_partition_id,
 543                                ioerr->oer_component.oid_object_id,
 544                                ioerr->oer_comp_offset,
 545                                ioerr->oer_comp_length);
 546
 547                        p = pnfs_osd_xdr_ioerr_reserve_space(xdr);
 548                        if (unlikely(!p)) {
 549                                res = -E2BIG;
 550                                break; /* accumulated_error */
 551                        }
 552
 553                        last_xdr = p;
 554                        pnfs_osd_xdr_encode_ioerr(p, &oir->ioerrs[i]);
 555                }
 556
 557                /* TODO: use xdr_write_pages */
 558                if (unlikely(res)) {
 559                        /* no space for even one error descriptor */
 560                        BUG_ON(!last_xdr);
 561
 562                        /* we've encountered a situation with lots and lots of
 563                         * errors and no space to encode them all. Use the last
 564                         * available slot to report the union of all the
 565                         * remaining errors.
 566                         */
 567                        encode_accumulated_error(objlay, last_xdr);
 568                        goto loop_done;
 569                }
 570                list_del(&oir->err_list);
 571                objio_free_result(oir);
 572        }
 573loop_done:
 574        spin_unlock(&objlay->lock);
 575
 576        *start = cpu_to_be32((xdr->p - start - 1) * 4);
 577        dprintk("%s: Return\n", __func__);
 578}
 579
 580
 581/*
 582 * Get Device Info API for io engines
 583 */
 584struct objlayout_deviceinfo {
 585        struct page *page;
 586        struct pnfs_osd_deviceaddr da; /* This must be last */
 587};
 588
 589/* Initialize and call nfs_getdeviceinfo, then decode and return a
 590 * "struct pnfs_osd_deviceaddr *" Eventually objlayout_put_deviceinfo()
 591 * should be called.
 592 */
 593int objlayout_get_deviceinfo(struct pnfs_layout_hdr *pnfslay,
 594        struct nfs4_deviceid *d_id, struct pnfs_osd_deviceaddr **deviceaddr,
 595        gfp_t gfp_flags)
 596{
 597        struct objlayout_deviceinfo *odi;
 598        struct pnfs_device pd;
 599        struct page *page, **pages;
 600        u32 *p;
 601        int err;
 602
 603        page = alloc_page(gfp_flags);
 604        if (!page)
 605                return -ENOMEM;
 606
 607        pages = &page;
 608        pd.pages = pages;
 609
 610        memcpy(&pd.dev_id, d_id, sizeof(*d_id));
 611        pd.layout_type = LAYOUT_OSD2_OBJECTS;
 612        pd.pages = &page;
 613        pd.pgbase = 0;
 614        pd.pglen = PAGE_SIZE;
 615        pd.mincount = 0;
 616
 617        err = nfs4_proc_getdeviceinfo(NFS_SERVER(pnfslay->plh_inode), &pd);
 618        dprintk("%s nfs_getdeviceinfo returned %d\n", __func__, err);
 619        if (err)
 620                goto err_out;
 621
 622        p = page_address(page);
 623        odi = kzalloc(sizeof(*odi), gfp_flags);
 624        if (!odi) {
 625                err = -ENOMEM;
 626                goto err_out;
 627        }
 628        pnfs_osd_xdr_decode_deviceaddr(&odi->da, p);
 629        odi->page = page;
 630        *deviceaddr = &odi->da;
 631        return 0;
 632
 633err_out:
 634        __free_page(page);
 635        return err;
 636}
 637
 638void objlayout_put_deviceinfo(struct pnfs_osd_deviceaddr *deviceaddr)
 639{
 640        struct objlayout_deviceinfo *odi = container_of(deviceaddr,
 641                                                struct objlayout_deviceinfo,
 642                                                da);
 643
 644        __free_page(odi->page);
 645        kfree(odi);
 646}
 647
 648enum {
 649        OBJLAYOUT_MAX_URI_LEN = 256, OBJLAYOUT_MAX_OSDNAME_LEN = 64,
 650        OBJLAYOUT_MAX_SYSID_HEX_LEN = OSD_SYSTEMID_LEN * 2 + 1,
 651        OSD_LOGIN_UPCALL_PATHLEN  = 256
 652};
 653
 654static char osd_login_prog[OSD_LOGIN_UPCALL_PATHLEN] = "/sbin/osd_login";
 655
 656module_param_string(osd_login_prog, osd_login_prog, sizeof(osd_login_prog),
 657                    0600);
 658MODULE_PARM_DESC(osd_login_prog, "Path to the osd_login upcall program");
 659
 660struct __auto_login {
 661        char uri[OBJLAYOUT_MAX_URI_LEN];
 662        char osdname[OBJLAYOUT_MAX_OSDNAME_LEN];
 663        char systemid_hex[OBJLAYOUT_MAX_SYSID_HEX_LEN];
 664};
 665
 666static int __objlayout_upcall(struct __auto_login *login)
 667{
 668        static char *envp[] = { "HOME=/",
 669                "TERM=linux",
 670                "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 671                NULL
 672        };
 673        char *argv[8];
 674        int ret;
 675
 676        if (unlikely(!osd_login_prog[0])) {
 677                dprintk("%s: osd_login_prog is disabled\n", __func__);
 678                return -EACCES;
 679        }
 680
 681        dprintk("%s uri: %s\n", __func__, login->uri);
 682        dprintk("%s osdname %s\n", __func__, login->osdname);
 683        dprintk("%s systemid_hex %s\n", __func__, login->systemid_hex);
 684
 685        argv[0] = (char *)osd_login_prog;
 686        argv[1] = "-u";
 687        argv[2] = login->uri;
 688        argv[3] = "-o";
 689        argv[4] = login->osdname;
 690        argv[5] = "-s";
 691        argv[6] = login->systemid_hex;
 692        argv[7] = NULL;
 693
 694        ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
 695        /*
 696         * Disable the upcall mechanism if we're getting an ENOENT or
 697         * EACCES error. The admin can re-enable it on the fly by using
 698         * sysfs to set the objlayoutdriver.osd_login_prog module parameter once
 699         * the problem has been fixed.
 700         */
 701        if (ret == -ENOENT || ret == -EACCES) {
 702                printk(KERN_ERR "PNFS-OBJ: %s was not found please set "
 703                        "objlayoutdriver.osd_login_prog kernel parameter!\n",
 704                        osd_login_prog);
 705                osd_login_prog[0] = '\0';
 706        }
 707        dprintk("%s %s return value: %d\n", __func__, osd_login_prog, ret);
 708
 709        return ret;
 710}
 711
 712/* Assume dest is all zeros */
 713static void __copy_nfsS_and_zero_terminate(struct nfs4_string s,
 714                                           char *dest, int max_len,
 715                                           const char *var_name)
 716{
 717        if (!s.len)
 718                return;
 719
 720        if (s.len >= max_len) {
 721                pr_warn_ratelimited(
 722                        "objlayout_autologin: %s: s.len(%d) >= max_len(%d)",
 723                        var_name, s.len, max_len);
 724                s.len = max_len - 1; /* space for null terminator */
 725        }
 726
 727        memcpy(dest, s.data, s.len);
 728}
 729
 730/* Assume sysid is all zeros */
 731static void _sysid_2_hex(struct nfs4_string s,
 732                  char sysid[OBJLAYOUT_MAX_SYSID_HEX_LEN])
 733{
 734        int i;
 735        char *cur;
 736
 737        if (!s.len)
 738                return;
 739
 740        if (s.len != OSD_SYSTEMID_LEN) {
 741                pr_warn_ratelimited(
 742                    "objlayout_autologin: systemid_len(%d) != OSD_SYSTEMID_LEN",
 743                    s.len);
 744                if (s.len > OSD_SYSTEMID_LEN)
 745                        s.len = OSD_SYSTEMID_LEN;
 746        }
 747
 748        cur = sysid;
 749        for (i = 0; i < s.len; i++)
 750                cur = hex_byte_pack(cur, s.data[i]);
 751}
 752
 753int objlayout_autologin(struct pnfs_osd_deviceaddr *deviceaddr)
 754{
 755        int rc;
 756        struct __auto_login login;
 757
 758        if (!deviceaddr->oda_targetaddr.ota_netaddr.r_addr.len)
 759                return -ENODEV;
 760
 761        memset(&login, 0, sizeof(login));
 762        __copy_nfsS_and_zero_terminate(
 763                deviceaddr->oda_targetaddr.ota_netaddr.r_addr,
 764                login.uri, sizeof(login.uri), "URI");
 765
 766        __copy_nfsS_and_zero_terminate(
 767                deviceaddr->oda_osdname,
 768                login.osdname, sizeof(login.osdname), "OSDNAME");
 769
 770        _sysid_2_hex(deviceaddr->oda_systemid, login.systemid_hex);
 771
 772        rc = __objlayout_upcall(&login);
 773        if (rc > 0) /* script returns positive values */
 774                rc = -ENODEV;
 775
 776        return rc;
 777}
 778