nfs/filelayout/filelayoutdev.c

0001 /*
0002  *  Device operations for the pnfs nfs4 file layout driver.
0003  *
0004  *  Copyright (c) 2002
0005  *  The Regents of the University of Michigan
0006  *  All Rights Reserved
0007  *
0008  *  Dean Hildebrand <dhildebz@umich.edu>
0009  *  Garth Goodson   <Garth.Goodson@netapp.com>
0010  *
0011  *  Permission is granted to use, copy, create derivative works, and
0012  *  redistribute this software and such derivative works for any purpose,
0013  *  so long as the name of the University of Michigan is not used in
0014  *  any advertising or publicity pertaining to the use or distribution
0015  *  of this software without specific, written prior authorization. If
0016  *  the above copyright notice or any other identification of the
0017  *  University of Michigan is included in any copy of any portion of
0018  *  this software, then the disclaimer below must also be included.
0019  *
0020  *  This software is provided as is, without representation or warranty
0021  *  of any kind either express or implied, including without limitation
0022  *  the implied warranties of merchantability, fitness for a particular
0023  *  purpose, or noninfringement.  The Regents of the University of
0024  *  Michigan shall not be liable for any damages, including special,
0025  *  indirect, incidental, or consequential damages, with respect to any
0026  *  claim arising out of or in connection with the use of the software,
0027  *  even if it has been or is hereafter advised of the possibility of
0028  *  such damages.
0029  */
0030
0031 #include <linux/nfs_fs.h>
0032 #include <linux/vmalloc.h>
0033 #include <linux/module.h>
0034
0035 #include "../internal.h"
0036 #include "../nfs4session.h"
0037 #include "filelayout.h"
0038
0039 #define NFSDBG_FACILITY     NFSDBG_PNFS_LD
0040
0041 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
0042 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
0043
0044 void
0045 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
0046 {
0047     struct nfs4_pnfs_ds *ds;
0048     int i;
0049
0050     nfs4_print_deviceid(&dsaddr->id_node.deviceid);
0051
0052     for (i = 0; i < dsaddr->ds_num; i++) {
0053         ds = dsaddr->ds_list[i];
0054         if (ds != NULL)
0055             nfs4_pnfs_ds_put(ds);
0056     }
0057     kfree(dsaddr->stripe_indices);
0058     kfree_rcu(dsaddr, id_node.rcu);
0059 }
0060
0061 /* Decode opaque device data and return the result */
0062 struct nfs4_file_layout_dsaddr *
0063 nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
0064         gfp_t gfp_flags)
0065 {
0066     int i;
0067     u32 cnt, num;
0068     u8 *indexp;
0069     __be32 *p;
0070     u8 *stripe_indices;
0071     u8 max_stripe_index;
0072     struct nfs4_file_layout_dsaddr *dsaddr = NULL;
0073     struct xdr_stream stream;
0074     struct xdr_buf buf;
0075     struct page *scratch;
0076     struct list_head dsaddrs;
0077     struct nfs4_pnfs_ds_addr *da;
0078
0079     /* set up xdr stream */
0080     scratch = alloc_page(gfp_flags);
0081     if (!scratch)
0082         goto out_err;
0083
0084     xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
0085     xdr_set_scratch_page(&stream, scratch);
0086
0087     /* Get the stripe count (number of stripe index) */
0088     p = xdr_inline_decode(&stream, 4);
0089     if (unlikely(!p))
0090         goto out_err_free_scratch;
0091
0092     cnt = be32_to_cpup(p);
0093     dprintk("%s stripe count  %d\n", __func__, cnt);
0094     if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
0095         printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
0096                "supported maximum %d\n", __func__,
0097             cnt, NFS4_PNFS_MAX_STRIPE_CNT);
0098         goto out_err_free_scratch;
0099     }
0100
0101     /* read stripe indices */
0102     stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
0103     if (!stripe_indices)
0104         goto out_err_free_scratch;
0105
0106     p = xdr_inline_decode(&stream, cnt << 2);
0107     if (unlikely(!p))
0108         goto out_err_free_stripe_indices;
0109
0110     indexp = &stripe_indices[0];
0111     max_stripe_index = 0;
0112     for (i = 0; i < cnt; i++) {
0113         *indexp = be32_to_cpup(p++);
0114         max_stripe_index = max(max_stripe_index, *indexp);
0115         indexp++;
0116     }
0117
0118     /* Check the multipath list count */
0119     p = xdr_inline_decode(&stream, 4);
0120     if (unlikely(!p))
0121         goto out_err_free_stripe_indices;
0122
0123     num = be32_to_cpup(p);
0124     dprintk("%s ds_num %u\n", __func__, num);
0125     if (num > NFS4_PNFS_MAX_MULTI_CNT) {
0126         printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
0127             "supported maximum %d\n", __func__,
0128             num, NFS4_PNFS_MAX_MULTI_CNT);
0129         goto out_err_free_stripe_indices;
0130     }
0131
0132     /* validate stripe indices are all < num */
0133     if (max_stripe_index >= num) {
0134         printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
0135             __func__, max_stripe_index, num);
0136         goto out_err_free_stripe_indices;
0137     }
0138
0139     dsaddr = kzalloc(struct_size(dsaddr, ds_list, num), gfp_flags);
0140     if (!dsaddr)
0141         goto out_err_free_stripe_indices;
0142
0143     dsaddr->stripe_count = cnt;
0144     dsaddr->stripe_indices = stripe_indices;
0145     stripe_indices = NULL;
0146     dsaddr->ds_num = num;
0147     nfs4_init_deviceid_node(&dsaddr->id_node, server, &pdev->dev_id);
0148
0149     INIT_LIST_HEAD(&dsaddrs);
0150
0151     for (i = 0; i < dsaddr->ds_num; i++) {
0152         int j;
0153         u32 mp_count;
0154
0155         p = xdr_inline_decode(&stream, 4);
0156         if (unlikely(!p))
0157             goto out_err_free_deviceid;
0158
0159         mp_count = be32_to_cpup(p); /* multipath count */
0160         for (j = 0; j < mp_count; j++) {
0161             da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
0162                             &stream, gfp_flags);
0163             if (da)
0164                 list_add_tail(&da->da_node, &dsaddrs);
0165         }
0166         if (list_empty(&dsaddrs)) {
0167             dprintk("%s: no suitable DS addresses found\n",
0168                 __func__);
0169             goto out_err_free_deviceid;
0170         }
0171
0172         dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
0173         if (!dsaddr->ds_list[i])
0174             goto out_err_drain_dsaddrs;
0175
0176         /* If DS was already in cache, free ds addrs */
0177         while (!list_empty(&dsaddrs)) {
0178             da = list_first_entry(&dsaddrs,
0179                           struct nfs4_pnfs_ds_addr,
0180                           da_node);
0181             list_del_init(&da->da_node);
0182             kfree(da->da_remotestr);
0183             kfree(da);
0184         }
0185     }
0186
0187     __free_page(scratch);
0188     return dsaddr;
0189
0190 out_err_drain_dsaddrs:
0191     while (!list_empty(&dsaddrs)) {
0192         da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
0193                       da_node);
0194         list_del_init(&da->da_node);
0195         kfree(da->da_remotestr);
0196         kfree(da);
0197     }
0198 out_err_free_deviceid:
0199     nfs4_fl_free_deviceid(dsaddr);
0200     /* stripe_indicies was part of dsaddr */
0201     goto out_err_free_scratch;
0202 out_err_free_stripe_indices:
0203     kfree(stripe_indices);
0204 out_err_free_scratch:
0205     __free_page(scratch);
0206 out_err:
0207     dprintk("%s ERROR: returning NULL\n", __func__);
0208     return NULL;
0209 }
0210
0211 void
0212 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
0213 {
0214     nfs4_put_deviceid_node(&dsaddr->id_node);
0215 }
0216
0217 /*
0218  * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
0219  * Then: ((res + fsi) % dsaddr->stripe_count)
0220  */
0221 u32
0222 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
0223 {
0224     struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
0225     u64 tmp;
0226
0227     tmp = offset - flseg->pattern_offset;
0228     do_div(tmp, flseg->stripe_unit);
0229     tmp += flseg->first_stripe_index;
0230     return do_div(tmp, flseg->dsaddr->stripe_count);
0231 }
0232
0233 u32
0234 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
0235 {
0236     return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
0237 }
0238
0239 struct nfs_fh *
0240 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
0241 {
0242     struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
0243     u32 i;
0244
0245     if (flseg->stripe_type == STRIPE_SPARSE) {
0246         if (flseg->num_fh == 1)
0247             i = 0;
0248         else if (flseg->num_fh == 0)
0249             /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
0250             return NULL;
0251         else
0252             i = nfs4_fl_calc_ds_index(lseg, j);
0253     } else
0254         i = j;
0255     return flseg->fh_array[i];
0256 }
0257
0258 /* Upon return, either ds is connected, or ds is NULL */
0259 struct nfs4_pnfs_ds *
0260 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
0261 {
0262     struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
0263     struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
0264     struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
0265     struct nfs4_pnfs_ds *ret = ds;
0266     struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
0267     int status;
0268
0269     if (ds == NULL) {
0270         printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
0271             __func__, ds_idx);
0272         pnfs_generic_mark_devid_invalid(devid);
0273         goto out;
0274     }
0275     smp_rmb();
0276     if (ds->ds_clp)
0277         goto out_test_devid;
0278
0279     status = nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
0280                  dataserver_retrans, 4,
0281                  s->nfs_client->cl_minorversion);
0282     if (status) {
0283         nfs4_mark_deviceid_unavailable(devid);
0284         ret = NULL;
0285         goto out;
0286     }
0287
0288 out_test_devid:
0289     if (ret->ds_clp == NULL ||
0290         filelayout_test_devid_unavailable(devid))
0291         ret = NULL;
0292 out:
0293     return ret;
0294 }
0295
0296 module_param(dataserver_retrans, uint, 0644);
0297 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
0298             "retries a request before it attempts further "
0299             " recovery  action.");
0300 module_param(dataserver_timeo, uint, 0644);
0301 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
0302             "NFSv4.1  client  waits for a response from a "
0303             " data server before it retries an NFS request.");