1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/nfs/read.c 4 * 5 * Block I/O for NFS 6 * 7 * Partial copy of Linus' read cache modifications to fs/nfs/file.c 8 * modified for async RPC by okir@monad.swb.de 9 */ 10 11 #include <linux/time.h> 12 #include <linux/kernel.h> 13 #include <linux/errno.h> 14 #include <linux/fcntl.h> 15 #include <linux/stat.h> 16 #include <linux/mm.h> 17 #include <linux/slab.h> 18 #include <linux/pagemap.h> 19 #include <linux/sunrpc/clnt.h> 20 #include <linux/nfs_fs.h> 21 #include <linux/nfs_page.h> 22 #include <linux/module.h> 23 24 #include "nfs4_fs.h" 25 #include "internal.h" 26 #include "iostat.h" 27 #include "fscache.h" 28 #include "pnfs.h" 29 #include "nfstrace.h" 30 31 #define NFSDBG_FACILITY NFSDBG_PAGECACHE 32 33 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops; 34 static const struct nfs_rw_ops nfs_rw_read_ops; 35 36 static struct kmem_cache *nfs_rdata_cachep; 37 38 static struct nfs_pgio_header *nfs_readhdr_alloc(void) 39 { 40 struct nfs_pgio_header *p = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL); 41 42 if (p) 43 p->rw_mode = FMODE_READ; 44 return p; 45 } 46 47 static void nfs_readhdr_free(struct nfs_pgio_header *rhdr) 48 { 49 kmem_cache_free(nfs_rdata_cachep, rhdr); 50 } 51 52 static 53 int nfs_return_empty_page(struct page *page) 54 { 55 zero_user(page, 0, PAGE_SIZE); 56 SetPageUptodate(page); 57 unlock_page(page); 58 return 0; 59 } 60 61 void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, 62 struct inode *inode, bool force_mds, 63 const struct nfs_pgio_completion_ops *compl_ops) 64 { 65 struct nfs_server *server = NFS_SERVER(inode); 66 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 67 68 #ifdef CONFIG_NFS_V4_1 69 if (server->pnfs_curr_ld && !force_mds) 70 pg_ops = server->pnfs_curr_ld->pg_read_ops; 71 #endif 72 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops, 73 server->rsize, 0); 74 } 75 EXPORT_SYMBOL_GPL(nfs_pageio_init_read); 76 77 static void nfs_pageio_complete_read(struct nfs_pageio_descriptor *pgio) 78 { 79 struct nfs_pgio_mirror *pgm; 80 unsigned long npages; 81 82 nfs_pageio_complete(pgio); 83 84 /* It doesn't make sense to do mirrored reads! */ 85 WARN_ON_ONCE(pgio->pg_mirror_count != 1); 86 87 pgm = &pgio->pg_mirrors[0]; 88 NFS_I(pgio->pg_inode)->read_io += pgm->pg_bytes_written; 89 npages = (pgm->pg_bytes_written + PAGE_SIZE - 1) >> PAGE_SHIFT; 90 nfs_add_stats(pgio->pg_inode, NFSIOS_READPAGES, npages); 91 } 92 93 94 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio) 95 { 96 struct nfs_pgio_mirror *mirror; 97 98 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 99 pgio->pg_ops->pg_cleanup(pgio); 100 101 pgio->pg_ops = &nfs_pgio_rw_ops; 102 103 /* read path should never have more than one mirror */ 104 WARN_ON_ONCE(pgio->pg_mirror_count != 1); 105 106 mirror = &pgio->pg_mirrors[0]; 107 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize; 108 } 109 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds); 110 111 static void nfs_readpage_release(struct nfs_page *req, int error) 112 { 113 struct inode *inode = d_inode(nfs_req_openctx(req)->dentry); 114 struct page *page = req->wb_page; 115 116 dprintk("NFS: read done (%s/%llu %d@%lld)\n", inode->i_sb->s_id, 117 (unsigned long long)NFS_FILEID(inode), req->wb_bytes, 118 (long long)req_offset(req)); 119 120 if (nfs_error_is_fatal_on_server(error) && error != -ETIMEDOUT) 121 SetPageError(page); 122 if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) { 123 struct address_space *mapping = page_file_mapping(page); 124 125 if (PageUptodate(page)) 126 nfs_fscache_write_page(inode, page); 127 else if (!PageError(page) && !PagePrivate(page)) 128 generic_error_remove_page(mapping, page); 129 unlock_page(page); 130 } 131 nfs_release_request(req); 132 } 133 134 struct nfs_readdesc { 135 struct nfs_pageio_descriptor pgio; 136 struct nfs_open_context *ctx; 137 }; 138 139 static void nfs_page_group_set_uptodate(struct nfs_page *req) 140 { 141 if (nfs_page_group_sync_on_bit(req, PG_UPTODATE)) 142 SetPageUptodate(req->wb_page); 143 } 144 145 static void nfs_read_completion(struct nfs_pgio_header *hdr) 146 { 147 unsigned long bytes = 0; 148 int error; 149 150 if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 151 goto out; 152 while (!list_empty(&hdr->pages)) { 153 struct nfs_page *req = nfs_list_entry(hdr->pages.next); 154 struct page *page = req->wb_page; 155 unsigned long start = req->wb_pgbase; 156 unsigned long end = req->wb_pgbase + req->wb_bytes; 157 158 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) { 159 /* note: regions of the page not covered by a 160 * request are zeroed in readpage_async_filler */ 161 if (bytes > hdr->good_bytes) { 162 /* nothing in this request was good, so zero 163 * the full extent of the request */ 164 zero_user_segment(page, start, end); 165 166 } else if (hdr->good_bytes - bytes < req->wb_bytes) { 167 /* part of this request has good bytes, but 168 * not all. zero the bad bytes */ 169 start += hdr->good_bytes - bytes; 170 WARN_ON(start < req->wb_pgbase); 171 zero_user_segment(page, start, end); 172 } 173 } 174 error = 0; 175 bytes += req->wb_bytes; 176 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { 177 if (bytes <= hdr->good_bytes) 178 nfs_page_group_set_uptodate(req); 179 else { 180 error = hdr->error; 181 xchg(&nfs_req_openctx(req)->error, error); 182 } 183 } else 184 nfs_page_group_set_uptodate(req); 185 nfs_list_remove_request(req); 186 nfs_readpage_release(req, error); 187 } 188 out: 189 hdr->release(hdr); 190 } 191 192 static void nfs_initiate_read(struct nfs_pgio_header *hdr, 193 struct rpc_message *msg, 194 const struct nfs_rpc_ops *rpc_ops, 195 struct rpc_task_setup *task_setup_data, int how) 196 { 197 rpc_ops->read_setup(hdr, msg); 198 trace_nfs_initiate_read(hdr); 199 } 200 201 static void 202 nfs_async_read_error(struct list_head *head, int error) 203 { 204 struct nfs_page *req; 205 206 while (!list_empty(head)) { 207 req = nfs_list_entry(head->next); 208 nfs_list_remove_request(req); 209 nfs_readpage_release(req, error); 210 } 211 } 212 213 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = { 214 .error_cleanup = nfs_async_read_error, 215 .completion = nfs_read_completion, 216 }; 217 218 /* 219 * This is the callback from RPC telling us whether a reply was 220 * received or some error occurred (timeout or socket shutdown). 221 */ 222 static int nfs_readpage_done(struct rpc_task *task, 223 struct nfs_pgio_header *hdr, 224 struct inode *inode) 225 { 226 int status = NFS_PROTO(inode)->read_done(task, hdr); 227 if (status != 0) 228 return status; 229 230 nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count); 231 trace_nfs_readpage_done(task, hdr); 232 233 if (task->tk_status == -ESTALE) { 234 nfs_set_inode_stale(inode); 235 nfs_mark_for_revalidate(inode); 236 } 237 return 0; 238 } 239 240 static void nfs_readpage_retry(struct rpc_task *task, 241 struct nfs_pgio_header *hdr) 242 { 243 struct nfs_pgio_args *argp = &hdr->args; 244 struct nfs_pgio_res *resp = &hdr->res; 245 246 /* This is a short read! */ 247 nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD); 248 trace_nfs_readpage_short(task, hdr); 249 250 /* Has the server at least made some progress? */ 251 if (resp->count == 0) { 252 nfs_set_pgio_error(hdr, -EIO, argp->offset); 253 return; 254 } 255 256 /* For non rpc-based layout drivers, retry-through-MDS */ 257 if (!task->tk_ops) { 258 hdr->pnfs_error = -EAGAIN; 259 return; 260 } 261 262 /* Yes, so retry the read at the end of the hdr */ 263 hdr->mds_offset += resp->count; 264 argp->offset += resp->count; 265 argp->pgbase += resp->count; 266 argp->count -= resp->count; 267 resp->count = 0; 268 resp->eof = 0; 269 rpc_restart_call_prepare(task); 270 } 271 272 static void nfs_readpage_result(struct rpc_task *task, 273 struct nfs_pgio_header *hdr) 274 { 275 if (hdr->res.eof) { 276 loff_t pos = hdr->args.offset + hdr->res.count; 277 unsigned int new = pos - hdr->io_start; 278 279 if (hdr->good_bytes > new) { 280 hdr->good_bytes = new; 281 set_bit(NFS_IOHDR_EOF, &hdr->flags); 282 clear_bit(NFS_IOHDR_ERROR, &hdr->flags); 283 } 284 } else if (hdr->res.count < hdr->args.count) 285 nfs_readpage_retry(task, hdr); 286 } 287 288 static int 289 readpage_async_filler(struct nfs_readdesc *desc, struct page *page) 290 { 291 struct inode *inode = page_file_mapping(page)->host; 292 unsigned int rsize = NFS_SERVER(inode)->rsize; 293 struct nfs_page *new; 294 unsigned int len, aligned_len; 295 int error; 296 297 len = nfs_page_length(page); 298 if (len == 0) 299 return nfs_return_empty_page(page); 300 301 aligned_len = min_t(unsigned int, ALIGN(len, rsize), PAGE_SIZE); 302 303 if (!IS_SYNC(page->mapping->host)) { 304 error = nfs_fscache_read_page(page->mapping->host, page); 305 if (error == 0) 306 goto out_unlock; 307 } 308 309 new = nfs_create_request(desc->ctx, page, 0, aligned_len); 310 if (IS_ERR(new)) 311 goto out_error; 312 313 if (len < PAGE_SIZE) 314 zero_user_segment(page, len, PAGE_SIZE); 315 if (!nfs_pageio_add_request(&desc->pgio, new)) { 316 nfs_list_remove_request(new); 317 error = desc->pgio.pg_error; 318 nfs_readpage_release(new, error); 319 goto out; 320 } 321 return 0; 322 out_error: 323 error = PTR_ERR(new); 324 out_unlock: 325 unlock_page(page); 326 out: 327 return error; 328 } 329 330 /* 331 * Read a page over NFS. 332 * We read the page synchronously in the following case: 333 * - The error flag is set for this page. This happens only when a 334 * previous async read operation failed. 335 */ 336 int nfs_readpage(struct file *file, struct page *page) 337 { 338 struct nfs_readdesc desc; 339 struct inode *inode = page_file_mapping(page)->host; 340 int ret; 341 342 trace_nfs_aop_readpage(inode, page); 343 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); 344 345 /* 346 * Try to flush any pending writes to the file.. 347 * 348 * NOTE! Because we own the page lock, there cannot 349 * be any new pending writes generated at this point 350 * for this page (other pages can be written to). 351 */ 352 ret = nfs_wb_page(inode, page); 353 if (ret) 354 goto out_unlock; 355 if (PageUptodate(page)) 356 goto out_unlock; 357 358 ret = -ESTALE; 359 if (NFS_STALE(inode)) 360 goto out_unlock; 361 362 if (file == NULL) { 363 ret = -EBADF; 364 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); 365 if (desc.ctx == NULL) 366 goto out_unlock; 367 } else 368 desc.ctx = get_nfs_open_context(nfs_file_open_context(file)); 369 370 xchg(&desc.ctx->error, 0); 371 nfs_pageio_init_read(&desc.pgio, inode, false, 372 &nfs_async_read_completion_ops); 373 374 ret = readpage_async_filler(&desc, page); 375 if (ret) 376 goto out; 377 378 nfs_pageio_complete_read(&desc.pgio); 379 ret = desc.pgio.pg_error < 0 ? desc.pgio.pg_error : 0; 380 if (!ret) { 381 ret = wait_on_page_locked_killable(page); 382 if (!PageUptodate(page) && !ret) 383 ret = xchg(&desc.ctx->error, 0); 384 } 385 out: 386 put_nfs_open_context(desc.ctx); 387 trace_nfs_aop_readpage_done(inode, page, ret); 388 return ret; 389 out_unlock: 390 unlock_page(page); 391 trace_nfs_aop_readpage_done(inode, page, ret); 392 return ret; 393 } 394 395 void nfs_readahead(struct readahead_control *ractl) 396 { 397 unsigned int nr_pages = readahead_count(ractl); 398 struct file *file = ractl->file; 399 struct nfs_readdesc desc; 400 struct inode *inode = ractl->mapping->host; 401 struct page *page; 402 int ret; 403 404 trace_nfs_aop_readahead(inode, readahead_pos(ractl), nr_pages); 405 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); 406 407 ret = -ESTALE; 408 if (NFS_STALE(inode)) 409 goto out; 410 411 if (file == NULL) { 412 ret = -EBADF; 413 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); 414 if (desc.ctx == NULL) 415 goto out; 416 } else 417 desc.ctx = get_nfs_open_context(nfs_file_open_context(file)); 418 419 nfs_pageio_init_read(&desc.pgio, inode, false, 420 &nfs_async_read_completion_ops); 421 422 while ((page = readahead_page(ractl)) != NULL) { 423 ret = readpage_async_filler(&desc, page); 424 put_page(page); 425 if (ret) 426 break; 427 } 428 429 nfs_pageio_complete_read(&desc.pgio); 430 431 put_nfs_open_context(desc.ctx); 432 out: 433 trace_nfs_aop_readahead_done(inode, nr_pages, ret); 434 } 435 436 int __init nfs_init_readpagecache(void) 437 { 438 nfs_rdata_cachep = kmem_cache_create("nfs_read_data", 439 sizeof(struct nfs_pgio_header), 440 0, SLAB_HWCACHE_ALIGN, 441 NULL); 442 if (nfs_rdata_cachep == NULL) 443 return -ENOMEM; 444 445 return 0; 446 } 447 448 void nfs_destroy_readpagecache(void) 449 { 450 kmem_cache_destroy(nfs_rdata_cachep); 451 } 452 453 static const struct nfs_rw_ops nfs_rw_read_ops = { 454 .rw_alloc_header = nfs_readhdr_alloc, 455 .rw_free_header = nfs_readhdr_free, 456 .rw_done = nfs_readpage_done, 457 .rw_result = nfs_readpage_result, 458 .rw_initiate = nfs_initiate_read, 459 }; 460