1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2014-2016 Christoph Hellwig. 4 */ 5 #include <linux/exportfs.h> 6 #include <linux/iomap.h> 7 #include <linux/slab.h> 8 #include <linux/pr.h> 9 10 #include <linux/nfsd/debug.h> 11 12 #include "blocklayoutxdr.h" 13 #include "pnfs.h" 14 #include "filecache.h" 15 #include "vfs.h" 16 17 #define NFSDDBG_FACILITY NFSDDBG_PNFS 18 19 20 static __be32 21 nfsd4_block_proc_layoutget(struct inode *inode, const struct svc_fh *fhp, 22 struct nfsd4_layoutget *args) 23 { 24 struct nfsd4_layout_seg *seg = &args->lg_seg; 25 struct super_block *sb = inode->i_sb; 26 u32 block_size = i_blocksize(inode); 27 struct pnfs_block_extent *bex; 28 struct iomap iomap; 29 u32 device_generation = 0; 30 int error; 31 32 if (seg->offset & (block_size - 1)) { 33 dprintk("pnfsd: I/O misaligned\n"); 34 goto out_layoutunavailable; 35 } 36 37 /* 38 * Some clients barf on non-zero block numbers for NONE or INVALID 39 * layouts, so make sure to zero the whole structure. 40 */ 41 error = -ENOMEM; 42 bex = kzalloc(sizeof(*bex), GFP_KERNEL); 43 if (!bex) 44 goto out_error; 45 args->lg_content = bex; 46 47 error = sb->s_export_op->map_blocks(inode, seg->offset, seg->length, 48 &iomap, seg->iomode != IOMODE_READ, 49 &device_generation); 50 if (error) { 51 if (error == -ENXIO) 52 goto out_layoutunavailable; 53 goto out_error; 54 } 55 56 if (iomap.length < args->lg_minlength) { 57 dprintk("pnfsd: extent smaller than minlength\n"); 58 goto out_layoutunavailable; 59 } 60 61 switch (iomap.type) { 62 case IOMAP_MAPPED: 63 if (seg->iomode == IOMODE_READ) 64 bex->es = PNFS_BLOCK_READ_DATA; 65 else 66 bex->es = PNFS_BLOCK_READWRITE_DATA; 67 bex->soff = iomap.addr; 68 break; 69 case IOMAP_UNWRITTEN: 70 if (seg->iomode & IOMODE_RW) { 71 /* 72 * Crack monkey special case from section 2.3.1. 73 */ 74 if (args->lg_minlength == 0) { 75 dprintk("pnfsd: no soup for you!\n"); 76 goto out_layoutunavailable; 77 } 78 79 bex->es = PNFS_BLOCK_INVALID_DATA; 80 bex->soff = iomap.addr; 81 break; 82 } 83 fallthrough; 84 case IOMAP_HOLE: 85 if (seg->iomode == IOMODE_READ) { 86 bex->es = PNFS_BLOCK_NONE_DATA; 87 break; 88 } 89 fallthrough; 90 case IOMAP_DELALLOC: 91 default: 92 WARN(1, "pnfsd: filesystem returned %d extent\n", iomap.type); 93 goto out_layoutunavailable; 94 } 95 96 error = nfsd4_set_deviceid(&bex->vol_id, fhp, device_generation); 97 if (error) 98 goto out_error; 99 bex->foff = iomap.offset; 100 bex->len = iomap.length; 101 102 seg->offset = iomap.offset; 103 seg->length = iomap.length; 104 105 dprintk("GET: 0x%llx:0x%llx %d\n", bex->foff, bex->len, bex->es); 106 return 0; 107 108 out_error: 109 seg->length = 0; 110 return nfserrno(error); 111 out_layoutunavailable: 112 seg->length = 0; 113 return nfserr_layoutunavailable; 114 } 115 116 static __be32 117 nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp, 118 struct iomap *iomaps, int nr_iomaps) 119 { 120 loff_t new_size = lcp->lc_last_wr + 1; 121 struct iattr iattr = { .ia_valid = 0 }; 122 int error; 123 124 if (lcp->lc_mtime.tv_nsec == UTIME_NOW || 125 timespec64_compare(&lcp->lc_mtime, &inode->i_mtime) < 0) 126 lcp->lc_mtime = current_time(inode); 127 iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME; 128 iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime; 129 130 if (new_size > i_size_read(inode)) { 131 iattr.ia_valid |= ATTR_SIZE; 132 iattr.ia_size = new_size; 133 } 134 135 error = inode->i_sb->s_export_op->commit_blocks(inode, iomaps, 136 nr_iomaps, &iattr); 137 kfree(iomaps); 138 return nfserrno(error); 139 } 140 141 #ifdef CONFIG_NFSD_BLOCKLAYOUT 142 static int 143 nfsd4_block_get_device_info_simple(struct super_block *sb, 144 struct nfsd4_getdeviceinfo *gdp) 145 { 146 struct pnfs_block_deviceaddr *dev; 147 struct pnfs_block_volume *b; 148 149 dev = kzalloc(sizeof(struct pnfs_block_deviceaddr) + 150 sizeof(struct pnfs_block_volume), GFP_KERNEL); 151 if (!dev) 152 return -ENOMEM; 153 gdp->gd_device = dev; 154 155 dev->nr_volumes = 1; 156 b = &dev->volumes[0]; 157 158 b->type = PNFS_BLOCK_VOLUME_SIMPLE; 159 b->simple.sig_len = PNFS_BLOCK_UUID_LEN; 160 return sb->s_export_op->get_uuid(sb, b->simple.sig, &b->simple.sig_len, 161 &b->simple.offset); 162 } 163 164 static __be32 165 nfsd4_block_proc_getdeviceinfo(struct super_block *sb, 166 struct svc_rqst *rqstp, 167 struct nfs4_client *clp, 168 struct nfsd4_getdeviceinfo *gdp) 169 { 170 if (bdev_is_partition(sb->s_bdev)) 171 return nfserr_inval; 172 return nfserrno(nfsd4_block_get_device_info_simple(sb, gdp)); 173 } 174 175 static __be32 176 nfsd4_block_proc_layoutcommit(struct inode *inode, 177 struct nfsd4_layoutcommit *lcp) 178 { 179 struct iomap *iomaps; 180 int nr_iomaps; 181 182 nr_iomaps = nfsd4_block_decode_layoutupdate(lcp->lc_up_layout, 183 lcp->lc_up_len, &iomaps, i_blocksize(inode)); 184 if (nr_iomaps < 0) 185 return nfserrno(nr_iomaps); 186 187 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps); 188 } 189 190 const struct nfsd4_layout_ops bl_layout_ops = { 191 /* 192 * Pretend that we send notification to the client. This is a blatant 193 * lie to force recent Linux clients to cache our device IDs. 194 * We rarely ever change the device ID, so the harm of leaking deviceids 195 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess 196 * in this regard, but I filed errata 4119 for this a while ago, and 197 * hopefully the Linux client will eventually start caching deviceids 198 * without this again. 199 */ 200 .notify_types = 201 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE, 202 .proc_getdeviceinfo = nfsd4_block_proc_getdeviceinfo, 203 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo, 204 .proc_layoutget = nfsd4_block_proc_layoutget, 205 .encode_layoutget = nfsd4_block_encode_layoutget, 206 .proc_layoutcommit = nfsd4_block_proc_layoutcommit, 207 }; 208 #endif /* CONFIG_NFSD_BLOCKLAYOUT */ 209 210 #ifdef CONFIG_NFSD_SCSILAYOUT 211 #define NFSD_MDS_PR_KEY 0x0100000000000000ULL 212 213 /* 214 * We use the client ID as a unique key for the reservations. 215 * This allows us to easily fence a client when recalls fail. 216 */ 217 static u64 nfsd4_scsi_pr_key(struct nfs4_client *clp) 218 { 219 return ((u64)clp->cl_clientid.cl_boot << 32) | clp->cl_clientid.cl_id; 220 } 221 222 static const u8 designator_types[] = { 223 PS_DESIGNATOR_EUI64, 224 PS_DESIGNATOR_NAA, 225 }; 226 227 static int 228 nfsd4_block_get_unique_id(struct gendisk *disk, struct pnfs_block_volume *b) 229 { 230 int ret, i; 231 232 for (i = 0; i < ARRAY_SIZE(designator_types); i++) { 233 u8 type = designator_types[i]; 234 235 ret = disk->fops->get_unique_id(disk, b->scsi.designator, type); 236 if (ret > 0) { 237 b->scsi.code_set = PS_CODE_SET_BINARY; 238 b->scsi.designator_type = type; 239 b->scsi.designator_len = ret; 240 return 0; 241 } 242 } 243 244 return -EINVAL; 245 } 246 247 static int 248 nfsd4_block_get_device_info_scsi(struct super_block *sb, 249 struct nfs4_client *clp, 250 struct nfsd4_getdeviceinfo *gdp) 251 { 252 struct pnfs_block_deviceaddr *dev; 253 struct pnfs_block_volume *b; 254 const struct pr_ops *ops; 255 int ret; 256 257 dev = kzalloc(sizeof(struct pnfs_block_deviceaddr) + 258 sizeof(struct pnfs_block_volume), GFP_KERNEL); 259 if (!dev) 260 return -ENOMEM; 261 gdp->gd_device = dev; 262 263 dev->nr_volumes = 1; 264 b = &dev->volumes[0]; 265 266 b->type = PNFS_BLOCK_VOLUME_SCSI; 267 b->scsi.pr_key = nfsd4_scsi_pr_key(clp); 268 269 ret = nfsd4_block_get_unique_id(sb->s_bdev->bd_disk, b); 270 if (ret < 0) 271 goto out_free_dev; 272 273 ret = -EINVAL; 274 ops = sb->s_bdev->bd_disk->fops->pr_ops; 275 if (!ops) { 276 pr_err("pNFS: device %s does not support PRs.\n", 277 sb->s_id); 278 goto out_free_dev; 279 } 280 281 ret = ops->pr_register(sb->s_bdev, 0, NFSD_MDS_PR_KEY, true); 282 if (ret) { 283 pr_err("pNFS: failed to register key for device %s.\n", 284 sb->s_id); 285 goto out_free_dev; 286 } 287 288 ret = ops->pr_reserve(sb->s_bdev, NFSD_MDS_PR_KEY, 289 PR_EXCLUSIVE_ACCESS_REG_ONLY, 0); 290 if (ret) { 291 pr_err("pNFS: failed to reserve device %s.\n", 292 sb->s_id); 293 goto out_free_dev; 294 } 295 296 return 0; 297 298 out_free_dev: 299 kfree(dev); 300 return ret; 301 } 302 303 static __be32 304 nfsd4_scsi_proc_getdeviceinfo(struct super_block *sb, 305 struct svc_rqst *rqstp, 306 struct nfs4_client *clp, 307 struct nfsd4_getdeviceinfo *gdp) 308 { 309 if (bdev_is_partition(sb->s_bdev)) 310 return nfserr_inval; 311 return nfserrno(nfsd4_block_get_device_info_scsi(sb, clp, gdp)); 312 } 313 static __be32 314 nfsd4_scsi_proc_layoutcommit(struct inode *inode, 315 struct nfsd4_layoutcommit *lcp) 316 { 317 struct iomap *iomaps; 318 int nr_iomaps; 319 320 nr_iomaps = nfsd4_scsi_decode_layoutupdate(lcp->lc_up_layout, 321 lcp->lc_up_len, &iomaps, i_blocksize(inode)); 322 if (nr_iomaps < 0) 323 return nfserrno(nr_iomaps); 324 325 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps); 326 } 327 328 static void 329 nfsd4_scsi_fence_client(struct nfs4_layout_stateid *ls) 330 { 331 struct nfs4_client *clp = ls->ls_stid.sc_client; 332 struct block_device *bdev = ls->ls_file->nf_file->f_path.mnt->mnt_sb->s_bdev; 333 334 bdev->bd_disk->fops->pr_ops->pr_preempt(bdev, NFSD_MDS_PR_KEY, 335 nfsd4_scsi_pr_key(clp), 0, true); 336 } 337 338 const struct nfsd4_layout_ops scsi_layout_ops = { 339 /* 340 * Pretend that we send notification to the client. This is a blatant 341 * lie to force recent Linux clients to cache our device IDs. 342 * We rarely ever change the device ID, so the harm of leaking deviceids 343 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess 344 * in this regard, but I filed errata 4119 for this a while ago, and 345 * hopefully the Linux client will eventually start caching deviceids 346 * without this again. 347 */ 348 .notify_types = 349 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE, 350 .proc_getdeviceinfo = nfsd4_scsi_proc_getdeviceinfo, 351 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo, 352 .proc_layoutget = nfsd4_block_proc_layoutget, 353 .encode_layoutget = nfsd4_block_encode_layoutget, 354 .proc_layoutcommit = nfsd4_scsi_proc_layoutcommit, 355 .fence_client = nfsd4_scsi_fence_client, 356 }; 357 #endif /* CONFIG_NFSD_SCSILAYOUT */ 358