1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* Network filesystem support services. 3 * 4 * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 * 7 * See: 8 * 9 * Documentation/filesystems/netfs_library.rst 10 * 11 * for a description of the network filesystem interface declared here. 12 */ 13 14 #ifndef _LINUX_NETFS_H 15 #define _LINUX_NETFS_H 16 17 #include <linux/workqueue.h> 18 #include <linux/fs.h> 19 #include <linux/pagemap.h> 20 #include <linux/uio.h> 21 #include <linux/rolling_buffer.h> 22 23 enum netfs_sreq_ref_trace; 24 typedef struct mempool mempool_t; 25 struct folio_queue; 26 27 /** 28 * folio_start_private_2 - Start an fscache write on a folio. [DEPRECATED] 29 * @folio: The folio. 30 * 31 * Call this function before writing a folio to a local cache. Starting a 32 * second write before the first one finishes is not allowed. 33 * 34 * Note that this should no longer be used. 35 */ 36 static inline void folio_start_private_2(struct folio *folio) 37 { 38 VM_BUG_ON_FOLIO(folio_test_private_2(folio), folio); 39 folio_get(folio); 40 folio_set_private_2(folio); 41 } 42 43 enum netfs_io_source { 44 NETFS_SOURCE_UNKNOWN, 45 NETFS_FILL_WITH_ZEROES, 46 NETFS_DOWNLOAD_FROM_SERVER, 47 NETFS_READ_FROM_CACHE, 48 NETFS_INVALID_READ, 49 NETFS_UPLOAD_TO_SERVER, 50 NETFS_WRITE_TO_CACHE, 51 } __mode(byte); 52 53 typedef void (*netfs_io_terminated_t)(void *priv, ssize_t transferred_or_error); 54 55 /* 56 * Per-inode context. This wraps the VFS inode. 57 */ 58 struct netfs_inode { 59 struct inode inode; /* The VFS inode */ 60 const struct netfs_request_ops *ops; 61 #if IS_ENABLED(CONFIG_FSCACHE) 62 struct fscache_cookie *cache; 63 #endif 64 struct mutex wb_lock; /* Writeback serialisation */ 65 loff_t _remote_i_size; /* Size of the remote file */ 66 loff_t _zero_point; /* Size after which we assume there's no data 67 * on the server */ 68 atomic_t io_count; /* Number of outstanding reqs */ 69 unsigned long flags; 70 #define NETFS_ICTX_ODIRECT 0 /* The file has DIO in progress */ 71 #define NETFS_ICTX_UNBUFFERED 1 /* I/O should not use the pagecache */ 72 #define NETFS_ICTX_MODIFIED_ATTR 3 /* Indicate change in mtime/ctime */ 73 #define NETFS_ICTX_SINGLE_NO_UPLOAD 4 /* Monolithic payload, cache but no upload */ 74 }; 75 76 /* 77 * A netfs group - for instance a ceph snap. This is marked on dirty pages and 78 * pages marked with a group must be flushed before they can be written under 79 * the domain of another group. 80 */ 81 struct netfs_group { 82 refcount_t ref; 83 void (*free)(struct netfs_group *netfs_group); 84 }; 85 86 /* 87 * Information about a dirty page (attached only if necessary). 88 * folio->private 89 */ 90 struct netfs_folio { 91 struct netfs_group *netfs_group; /* Filesystem's grouping marker (or NULL). */ 92 unsigned int dirty_offset; /* Write-streaming dirty data offset */ 93 unsigned int dirty_len; /* Write-streaming dirty data length */ 94 }; 95 #define NETFS_FOLIO_INFO 0x1UL /* OR'd with folio->private. */ 96 #define NETFS_FOLIO_COPY_TO_CACHE ((struct netfs_group *)0x356UL) /* Write to the cache only */ 97 98 static inline bool netfs_is_folio_info(const void *priv) 99 { 100 return (unsigned long)priv & NETFS_FOLIO_INFO; 101 } 102 103 static inline struct netfs_folio *__netfs_folio_info(const void *priv) 104 { 105 if (netfs_is_folio_info(priv)) 106 return (struct netfs_folio *)((unsigned long)priv & ~NETFS_FOLIO_INFO); 107 return NULL; 108 } 109 110 static inline struct netfs_folio *netfs_folio_info(struct folio *folio) 111 { 112 return __netfs_folio_info(folio_get_private(folio)); 113 } 114 115 static inline struct netfs_group *netfs_folio_group(struct folio *folio) 116 { 117 struct netfs_folio *finfo; 118 void *priv = folio_get_private(folio); 119 120 finfo = netfs_folio_info(folio); 121 if (finfo) 122 return finfo->netfs_group; 123 return priv; 124 } 125 126 /* 127 * Stream of I/O subrequests going to a particular destination, such as the 128 * server or the local cache. This is mainly intended for writing where we may 129 * have to write to multiple destinations concurrently. 130 */ 131 struct netfs_io_stream { 132 /* Submission tracking */ 133 struct netfs_io_subrequest *construct; /* Op being constructed */ 134 size_t sreq_max_len; /* Maximum size of a subrequest */ 135 unsigned int sreq_max_segs; /* 0 or max number of segments in an iterator */ 136 unsigned int submit_off; /* Folio offset we're submitting from */ 137 unsigned int submit_len; /* Amount of data left to submit */ 138 unsigned int submit_extendable_to; /* Amount I/O can be rounded up to */ 139 void (*prepare_write)(struct netfs_io_subrequest *subreq); 140 void (*issue_write)(struct netfs_io_subrequest *subreq); 141 /* Collection tracking */ 142 struct list_head subrequests; /* Contributory I/O operations */ 143 unsigned long long collected_to; /* Position we've collected results to */ 144 size_t transferred; /* The amount transferred from this stream */ 145 unsigned short error; /* Aggregate error for the stream */ 146 enum netfs_io_source source; /* Where to read from/write to */ 147 unsigned char stream_nr; /* Index of stream in parent table */ 148 bool avail; /* T if stream is available */ 149 bool active; /* T if stream is active */ 150 bool need_retry; /* T if this stream needs retrying */ 151 bool failed; /* T if this stream failed */ 152 bool transferred_valid; /* T is ->transferred is valid */ 153 }; 154 155 /* 156 * Resources required to do operations on a cache. 157 */ 158 struct netfs_cache_resources { 159 const struct netfs_cache_ops *ops; 160 void *cache_priv; 161 void *cache_priv2; 162 unsigned int debug_id; /* Cookie debug ID */ 163 unsigned int inval_counter; /* object->inval_counter at begin_op */ 164 }; 165 166 /* 167 * Descriptor for a single component subrequest. Each operation represents an 168 * individual read/write from/to a server, a cache, a journal, etc.. 169 * 170 * The buffer iterator is persistent for the life of the subrequest struct and 171 * the pages it points to can be relied on to exist for the duration. 172 */ 173 struct netfs_io_subrequest { 174 struct netfs_io_request *rreq; /* Supervising I/O request */ 175 struct work_struct work; 176 struct list_head rreq_link; /* Link in rreq->subrequests */ 177 struct iov_iter io_iter; /* Iterator for this subrequest */ 178 unsigned long long start; /* Where to start the I/O */ 179 size_t len; /* Size of the I/O */ 180 size_t transferred; /* Amount of data transferred */ 181 refcount_t ref; 182 short error; /* 0 or error that occurred */ 183 unsigned short debug_index; /* Index in list (for debugging output) */ 184 unsigned int nr_segs; /* Number of segs in io_iter */ 185 u8 retry_count; /* The number of retries (0 on initial pass) */ 186 enum netfs_io_source source; /* Where to read from/write to */ 187 unsigned char stream_nr; /* I/O stream this belongs to */ 188 unsigned long flags; 189 #define NETFS_SREQ_COPY_TO_CACHE 0 /* Set if should copy the data to the cache */ 190 #define NETFS_SREQ_CLEAR_TAIL 1 /* Set if the rest of the read should be cleared */ 191 #define NETFS_SREQ_MADE_PROGRESS 4 /* Set if we transferred at least some data */ 192 #define NETFS_SREQ_ONDEMAND 5 /* Set if it's from on-demand read mode */ 193 #define NETFS_SREQ_BOUNDARY 6 /* Set if ends on hard boundary (eg. ceph object) */ 194 #define NETFS_SREQ_HIT_EOF 7 /* Set if short due to EOF */ 195 #define NETFS_SREQ_IN_PROGRESS 8 /* Unlocked when the subrequest completes */ 196 #define NETFS_SREQ_NEED_RETRY 9 /* Set if the filesystem requests a retry */ 197 #define NETFS_SREQ_FAILED 10 /* Set if the subreq failed unretryably */ 198 }; 199 200 enum netfs_io_origin { 201 NETFS_READAHEAD, /* This read was triggered by readahead */ 202 NETFS_READPAGE, /* This read is a synchronous read */ 203 NETFS_READ_GAPS, /* This read is a synchronous read to fill gaps */ 204 NETFS_READ_SINGLE, /* This read should be treated as a single object */ 205 NETFS_READ_FOR_WRITE, /* This read is to prepare a write */ 206 NETFS_UNBUFFERED_READ, /* This is an unbuffered read */ 207 NETFS_DIO_READ, /* This is a direct I/O read */ 208 NETFS_WRITEBACK, /* This write was triggered by writepages */ 209 NETFS_WRITEBACK_SINGLE, /* This monolithic write was triggered by writepages */ 210 NETFS_WRITETHROUGH, /* This write was made by netfs_perform_write() */ 211 NETFS_UNBUFFERED_WRITE, /* This is an unbuffered write */ 212 NETFS_DIO_WRITE, /* This is a direct I/O write */ 213 NETFS_PGPRIV2_COPY_TO_CACHE, /* [DEPRECATED] This is writing read data to the cache */ 214 nr__netfs_io_origin 215 } __mode(byte); 216 217 /* 218 * Descriptor for an I/O helper request. This is used to make multiple I/O 219 * operations to a variety of data stores and then stitch the result together. 220 */ 221 struct netfs_io_request { 222 union { 223 struct work_struct cleanup_work; /* Deferred cleanup work */ 224 struct rcu_head rcu; 225 }; 226 struct work_struct work; /* Result collector work */ 227 struct inode *inode; /* The file being accessed */ 228 struct address_space *mapping; /* The mapping being accessed */ 229 struct kiocb *iocb; /* AIO completion vector */ 230 struct netfs_cache_resources cache_resources; 231 struct netfs_io_request *copy_to_cache; /* Request to write just-read data to the cache */ 232 #ifdef CONFIG_PROC_FS 233 struct list_head proc_link; /* Link in netfs_iorequests */ 234 #endif 235 struct netfs_io_stream io_streams[2]; /* Streams of parallel I/O operations */ 236 #define NR_IO_STREAMS 2 //wreq->nr_io_streams 237 struct netfs_group *group; /* Writeback group being written back */ 238 struct rolling_buffer buffer; /* Unencrypted buffer */ 239 #define NETFS_ROLLBUF_PUT_MARK ROLLBUF_MARK_1 240 #define NETFS_ROLLBUF_PAGECACHE_MARK ROLLBUF_MARK_2 241 wait_queue_head_t waitq; /* Processor waiter */ 242 void *netfs_priv; /* Private data for the netfs */ 243 void *netfs_priv2; /* Private data for the netfs */ 244 struct bio_vec *direct_bv; /* DIO buffer list (when handling iovec-iter) */ 245 unsigned long long submitted; /* Amount submitted for I/O so far */ 246 unsigned long long len; /* Length of the request */ 247 size_t transferred; /* Amount to be indicated as transferred */ 248 long error; /* 0 or error that occurred */ 249 unsigned long long i_size; /* Size of the file */ 250 unsigned long long start; /* Start position */ 251 atomic64_t issued_to; /* Write issuer folio cursor */ 252 unsigned long long collected_to; /* Point we've collected to */ 253 unsigned long long cleaned_to; /* Position we've cleaned folios to */ 254 unsigned long long abandon_to; /* Position to abandon folios to */ 255 const struct folio *no_unlock_folio; /* Don't unlock this folio after read */ 256 unsigned int direct_bv_count; /* Number of elements in direct_bv[] */ 257 unsigned int debug_id; 258 unsigned int rsize; /* Maximum read size (0 for none) */ 259 unsigned int wsize; /* Maximum write size (0 for none) */ 260 atomic_t subreq_counter; /* Next subreq->debug_index */ 261 unsigned int nr_group_rel; /* Number of refs to release on ->group */ 262 spinlock_t lock; /* Lock for queuing subreqs */ 263 unsigned char front_folio_order; /* Order (size) of front folio */ 264 enum netfs_io_origin origin; /* Origin of the request */ 265 bool direct_bv_unpin; /* T if direct_bv[] must be unpinned */ 266 refcount_t ref; 267 unsigned long flags; 268 #define NETFS_RREQ_IN_PROGRESS 0 /* Unlocked when the request completes (has ref) */ 269 #define NETFS_RREQ_ALL_QUEUED 1 /* All subreqs are now queued */ 270 #define NETFS_RREQ_PAUSE 2 /* Pause subrequest generation */ 271 #define NETFS_RREQ_FAILED 3 /* The request failed */ 272 #define NETFS_RREQ_RETRYING 4 /* Set if we're in the retry path */ 273 #define NETFS_RREQ_SHORT_TRANSFER 5 /* Set if we have a short transfer */ 274 #define NETFS_RREQ_OFFLOAD_COLLECTION 8 /* Offload collection to workqueue */ 275 #define NETFS_RREQ_NO_UNLOCK_FOLIO 9 /* Don't unlock no_unlock_folio on completion */ 276 #define NETFS_RREQ_FOLIO_COPY_TO_CACHE 10 /* Copy current folio to cache from read */ 277 #define NETFS_RREQ_UPLOAD_TO_SERVER 11 /* Need to write to the server */ 278 #define NETFS_RREQ_USE_IO_ITER 12 /* Use ->io_iter rather than ->i_pages */ 279 #define NETFS_RREQ_USE_PGPRIV2 31 /* [DEPRECATED] Use PG_private_2 to mark 280 * write to cache on read */ 281 const struct netfs_request_ops *netfs_ops; 282 }; 283 284 /* 285 * Operations the network filesystem can/must provide to the helpers. 286 */ 287 struct netfs_request_ops { 288 mempool_t *request_pool; 289 mempool_t *subrequest_pool; 290 int (*init_request)(struct netfs_io_request *rreq, struct file *file); 291 void (*free_request)(struct netfs_io_request *rreq); 292 void (*free_subrequest)(struct netfs_io_subrequest *rreq); 293 294 /* Read request handling */ 295 void (*expand_readahead)(struct netfs_io_request *rreq); 296 int (*prepare_read)(struct netfs_io_subrequest *subreq); 297 void (*issue_read)(struct netfs_io_subrequest *subreq); 298 bool (*is_still_valid)(struct netfs_io_request *rreq); 299 int (*check_write_begin)(struct file *file, loff_t pos, unsigned len, 300 struct folio **foliop, void **_fsdata); 301 void (*done)(struct netfs_io_request *rreq); 302 303 /* Modification handling */ 304 void (*update_i_size)(struct inode *inode, loff_t i_size); 305 void (*post_modify)(struct inode *inode); 306 307 /* Write request handling */ 308 void (*begin_writeback)(struct netfs_io_request *wreq); 309 void (*prepare_write)(struct netfs_io_subrequest *subreq); 310 void (*issue_write)(struct netfs_io_subrequest *subreq); 311 void (*retry_request)(struct netfs_io_request *wreq, struct netfs_io_stream *stream); 312 void (*invalidate_cache)(struct netfs_io_request *wreq); 313 }; 314 315 /* 316 * How to handle reading from a hole. 317 */ 318 enum netfs_read_from_hole { 319 NETFS_READ_HOLE_IGNORE, 320 NETFS_READ_HOLE_FAIL, 321 }; 322 323 /* 324 * Table of operations for access to a cache. 325 */ 326 struct netfs_cache_ops { 327 /* End an operation */ 328 void (*end_operation)(struct netfs_cache_resources *cres); 329 330 /* Read data from the cache */ 331 int (*read)(struct netfs_cache_resources *cres, 332 loff_t start_pos, 333 struct iov_iter *iter, 334 enum netfs_read_from_hole read_hole, 335 netfs_io_terminated_t term_func, 336 void *term_func_priv); 337 338 /* Write data to the cache */ 339 int (*write)(struct netfs_cache_resources *cres, 340 loff_t start_pos, 341 struct iov_iter *iter, 342 netfs_io_terminated_t term_func, 343 void *term_func_priv); 344 345 /* Write data to the cache from a netfs subrequest. */ 346 void (*issue_write)(struct netfs_io_subrequest *subreq); 347 348 /* Expand readahead request */ 349 void (*expand_readahead)(struct netfs_cache_resources *cres, 350 unsigned long long *_start, 351 unsigned long long *_len, 352 unsigned long long i_size); 353 354 /* Prepare a read operation, shortening it to a cached/uncached 355 * boundary as appropriate. 356 */ 357 enum netfs_io_source (*prepare_read)(struct netfs_io_subrequest *subreq, 358 unsigned long long i_size); 359 360 /* Prepare a write subrequest, working out if we're allowed to do it 361 * and finding out the maximum amount of data to gather before 362 * attempting to submit. If we're not permitted to do it, the 363 * subrequest should be marked failed. 364 */ 365 void (*prepare_write_subreq)(struct netfs_io_subrequest *subreq); 366 367 /* Prepare a write operation, working out what part of the write we can 368 * actually do. 369 */ 370 int (*prepare_write)(struct netfs_cache_resources *cres, 371 loff_t *_start, size_t *_len, size_t upper_len, 372 loff_t i_size, bool no_space_allocated_yet); 373 374 /* Prepare an on-demand read operation, shortening it to a cached/uncached 375 * boundary as appropriate. 376 */ 377 enum netfs_io_source (*prepare_ondemand_read)(struct netfs_cache_resources *cres, 378 loff_t start, size_t *_len, 379 loff_t i_size, 380 unsigned long *_flags, ino_t ino); 381 382 /* Query the occupancy of the cache in a region, returning where the 383 * next chunk of data starts and how long it is. 384 */ 385 int (*query_occupancy)(struct netfs_cache_resources *cres, 386 loff_t start, size_t len, size_t granularity, 387 loff_t *_data_start, size_t *_data_len); 388 }; 389 390 /* High-level read API. */ 391 ssize_t netfs_unbuffered_read_iter_locked(struct kiocb *iocb, struct iov_iter *iter); 392 ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter); 393 ssize_t netfs_buffered_read_iter(struct kiocb *iocb, struct iov_iter *iter); 394 ssize_t netfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter); 395 396 /* High-level write API */ 397 ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter, 398 struct netfs_group *netfs_group); 399 ssize_t netfs_buffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *from, 400 struct netfs_group *netfs_group); 401 ssize_t netfs_unbuffered_write_iter(struct kiocb *iocb, struct iov_iter *from); 402 ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *iter, 403 struct netfs_group *netfs_group); 404 ssize_t netfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from); 405 406 /* Single, monolithic object read/write API. */ 407 void netfs_single_mark_inode_dirty(struct inode *inode); 408 ssize_t netfs_read_single(struct inode *inode, struct file *file, struct iov_iter *iter); 409 int netfs_writeback_single(struct address_space *mapping, 410 struct writeback_control *wbc, 411 struct iov_iter *iter); 412 413 /* Address operations API */ 414 struct readahead_control; 415 void netfs_readahead(struct readahead_control *); 416 int netfs_read_folio(struct file *, struct folio *); 417 int netfs_write_begin(struct netfs_inode *, struct file *, 418 struct address_space *, loff_t pos, unsigned int len, 419 struct folio **, void **fsdata); 420 int netfs_writepages(struct address_space *mapping, 421 struct writeback_control *wbc); 422 bool netfs_dirty_folio(struct address_space *mapping, struct folio *folio); 423 int netfs_unpin_writeback(struct inode *inode, struct writeback_control *wbc); 424 void netfs_clear_inode_writeback(struct inode *inode, const void *aux); 425 void netfs_invalidate_folio(struct folio *folio, size_t offset, size_t length); 426 bool netfs_release_folio(struct folio *folio, gfp_t gfp); 427 428 /* VMA operations API. */ 429 vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group); 430 431 /* (Sub)request management API. */ 432 void netfs_read_subreq_progress(struct netfs_io_subrequest *subreq); 433 void netfs_read_subreq_terminated(struct netfs_io_subrequest *subreq); 434 void netfs_get_subrequest(struct netfs_io_subrequest *subreq, 435 enum netfs_sreq_ref_trace what); 436 void netfs_put_subrequest(struct netfs_io_subrequest *subreq, 437 enum netfs_sreq_ref_trace what); 438 ssize_t netfs_extract_user_iter(struct iov_iter *orig, size_t orig_len, 439 struct iov_iter *new, 440 iov_iter_extraction_t extraction_flags); 441 size_t netfs_limit_iter(const struct iov_iter *iter, size_t start_offset, 442 size_t max_size, size_t max_segs); 443 void netfs_prepare_write_failed(struct netfs_io_subrequest *subreq); 444 void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error); 445 446 int netfs_start_io_read(struct inode *inode); 447 void netfs_end_io_read(struct inode *inode); 448 int netfs_start_io_write(struct inode *inode); 449 void netfs_end_io_write(struct inode *inode); 450 int netfs_start_io_direct(struct inode *inode); 451 void netfs_end_io_direct(struct inode *inode); 452 453 /* Miscellaneous APIs. */ 454 struct folio_queue *netfs_folioq_alloc(unsigned int rreq_id, gfp_t gfp, 455 unsigned int trace /*enum netfs_folioq_trace*/); 456 void netfs_folioq_free(struct folio_queue *folioq, 457 unsigned int trace /*enum netfs_trace_folioq*/); 458 459 /* Buffer wrangling helpers API. */ 460 int netfs_alloc_folioq_buffer(struct address_space *mapping, 461 struct folio_queue **_buffer, 462 size_t *_cur_size, ssize_t size, gfp_t gfp); 463 void netfs_free_folioq_buffer(struct folio_queue *fq); 464 465 /** 466 * netfs_inode - Get the netfs inode context from the inode 467 * @inode: The inode to query 468 * 469 * Get the netfs lib inode context from the network filesystem's inode. The 470 * context struct is expected to directly follow on from the VFS inode struct. 471 */ 472 static inline struct netfs_inode *netfs_inode(struct inode *inode) 473 { 474 return container_of(inode, struct netfs_inode, inode); 475 } 476 477 /** 478 * netfs_read_remote_i_size - Read remote_i_size safely 479 * @inode: The inode to access 480 * 481 * Read remote_i_size safely without the potential for tearing on 32-bit 482 * arches. 483 * 484 * NOTE: in a 32bit arch with a preemptable kernel and an UP compile the 485 * i_size_read/write must be atomic with respect to the local cpu (unlike with 486 * preempt disabled), but they don't need to be atomic with respect to other 487 * cpus like in true SMP (so they need either to either locally disable irq 488 * around the read or for example on x86 they can be still implemented as a 489 * cmpxchg8b without the need of the lock prefix). For SMP compiles and 64bit 490 * archs it makes no difference if preempt is enabled or not. 491 */ 492 static inline unsigned long long netfs_read_remote_i_size(const struct inode *inode) 493 { 494 const struct netfs_inode *ictx = container_of(inode, struct netfs_inode, inode); 495 unsigned long long remote_i_size; 496 497 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 498 unsigned int seq; 499 500 do { 501 seq = read_seqcount_begin(&inode->i_size_seqcount); 502 remote_i_size = ictx->_remote_i_size; 503 } while (read_seqcount_retry(&inode->i_size_seqcount, seq)); 504 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 505 preempt_disable(); 506 remote_i_size = ictx->_remote_i_size; 507 preempt_enable(); 508 #else 509 /* Pairs with smp_store_release() in netfs_write_remote_i_size() */ 510 remote_i_size = smp_load_acquire(&ictx->_remote_i_size); 511 #endif 512 return remote_i_size; 513 } 514 515 /* 516 * netfs_write_remote_i_size - Set remote_i_size safely 517 * @inode: The inode to access 518 * @remote_i_size: The new value for the size of the file on the server 519 * 520 * Set remote_i_size safely without the potential for tearing on 32-bit arches. 521 * 522 * Context: The caller must hold inode->i_lock. 523 * 524 * NOTE: unlike netfs_read_remote_i_size(), netfs_write_remote_i_size() does 525 * need locking around it (normally i_rwsem), otherwise on 32bit/SMP an update 526 * of i_size_seqcount can be lost, resulting in subsequent i_size_read() calls 527 * spinning forever. 528 */ 529 static inline void netfs_write_remote_i_size(struct inode *inode, 530 unsigned long long remote_i_size) 531 { 532 struct netfs_inode *ictx = netfs_inode(inode); 533 534 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 535 write_seqcount_begin(&inode->i_size_seqcount); 536 ictx->_remote_i_size = remote_i_size; 537 write_seqcount_end(&inode->i_size_seqcount); 538 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 539 preempt_disable(); 540 ictx->_remote_i_size = remote_i_size; 541 preempt_enable(); 542 #else 543 /* 544 * Pairs with smp_load_acquire() in netfs_read_remote_i_size() to 545 * ensure changes related to inode size (such as page contents) are 546 * visible before we see the changed inode size. 547 */ 548 smp_store_release(&ictx->_remote_i_size, remote_i_size); 549 #endif 550 } 551 552 /** 553 * netfs_read_zero_point - Read zero_point safely 554 * @inode: The inode to access 555 * 556 * Read zero_point safely without the potential for tearing on 32-bit 557 * arches. 558 * 559 * NOTE: in a 32bit arch with a preemptable kernel and an UP compile the 560 * i_size_read/write must be atomic with respect to the local cpu (unlike with 561 * preempt disabled), but they don't need to be atomic with respect to other 562 * cpus like in true SMP (so they need either to either locally disable irq 563 * around the read or for example on x86 they can be still implemented as a 564 * cmpxchg8b without the need of the lock prefix). For SMP compiles and 64bit 565 * archs it makes no difference if preempt is enabled or not. 566 */ 567 static inline unsigned long long netfs_read_zero_point(const struct inode *inode) 568 { 569 struct netfs_inode *ictx = container_of(inode, struct netfs_inode, inode); 570 unsigned long long zero_point; 571 572 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 573 unsigned int seq; 574 575 do { 576 seq = read_seqcount_begin(&inode->i_size_seqcount); 577 zero_point = ictx->_zero_point; 578 } while (read_seqcount_retry(&inode->i_size_seqcount, seq)); 579 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 580 preempt_disable(); 581 zero_point = ictx->_zero_point; 582 preempt_enable(); 583 #else 584 /* Pairs with smp_store_release() in netfs_write_zero_point() */ 585 zero_point = smp_load_acquire(&ictx->_zero_point); 586 #endif 587 return zero_point; 588 } 589 590 /* 591 * netfs_write_zero_point - Set zero_point safely 592 * @inode: The inode to access 593 * @zero_point: The new value for the point beyond which the server has no data 594 * 595 * Set zero_point safely without the potential for tearing on 32-bit arches. 596 * 597 * Context: The caller must hold inode->i_lock. 598 * 599 * NOTE: unlike netfs_read_zero_point(), netfs_write_zero_point() does need 600 * locking around it (normally i_rwsem), otherwise on 32bit/SMP an update of 601 * i_size_seqcount can be lost, resulting in subsequent read calls spinning 602 * forever. 603 */ 604 static inline void netfs_write_zero_point(struct inode *inode, 605 unsigned long long zero_point) 606 { 607 struct netfs_inode *ictx = netfs_inode(inode); 608 609 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 610 write_seqcount_begin(&inode->i_size_seqcount); 611 ictx->_zero_point = zero_point; 612 write_seqcount_end(&inode->i_size_seqcount); 613 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 614 preempt_disable(); 615 ictx->_zero_point = zero_point; 616 preempt_enable(); 617 #else 618 /* 619 * Pairs with smp_load_acquire() in netfs_read_zero_point() to 620 * ensure changes related to inode size (such as page contents) are 621 * visible before we see the changed inode size. 622 */ 623 smp_store_release(&ictx->_zero_point, zero_point); 624 #endif 625 } 626 627 /** 628 * netfs_read_sizes - Read remote_i_size and zero_point safely 629 * @inode: The inode to access 630 * @i_size: Where to return the local file size. 631 * @remote_i_size: Where to return the size of the file on the server 632 * @zero_point: Where to return the the point beyond which the server has no data 633 * 634 * Read remote_i_size and zero_point safely without the potential for tearing 635 * on 32-bit arches. 636 * 637 * NOTE: in a 32bit arch with a preemptable kernel and an UP compile the 638 * i_size_read/write must be atomic with respect to the local cpu (unlike with 639 * preempt disabled), but they don't need to be atomic with respect to other 640 * cpus like in true SMP (so they need either to either locally disable irq 641 * around the read or for example on x86 they can be still implemented as a 642 * cmpxchg8b without the need of the lock prefix). For SMP compiles and 64bit 643 * archs it makes no difference if preempt is enabled or not. 644 */ 645 static inline void netfs_read_sizes(const struct inode *inode, 646 unsigned long long *i_size, 647 unsigned long long *remote_i_size, 648 unsigned long long *zero_point) 649 { 650 const struct netfs_inode *ictx = container_of(inode, struct netfs_inode, inode); 651 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 652 unsigned int seq; 653 654 do { 655 seq = read_seqcount_begin(&inode->i_size_seqcount); 656 *i_size = inode->i_size; 657 *remote_i_size = ictx->_remote_i_size; 658 *zero_point = ictx->_zero_point; 659 } while (read_seqcount_retry(&inode->i_size_seqcount, seq)); 660 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 661 preempt_disable(); 662 *i_size = inode->i_size; 663 *remote_i_size = ictx->_remote_i_size; 664 *zero_point = ictx->_zero_point; 665 preempt_enable(); 666 #else 667 /* Pairs with smp_store_release() in i_size_write() */ 668 *i_size = smp_load_acquire(&inode->i_size); 669 /* Pairs with smp_store_release() in netfs_write_remote_i_size() */ 670 *remote_i_size = smp_load_acquire(&ictx->_remote_i_size); 671 /* Pairs with smp_store_release() in netfs_write_zero_point() */ 672 *zero_point = smp_load_acquire(&ictx->_zero_point); 673 #endif 674 } 675 676 /* 677 * netfs_write_sizes - Set i_size, remote_i_size and zero_point safely 678 * @inode: The inode to access 679 * @i_size: The new value for the local size of the file 680 * @remote_i_size: The new value for the size of the file on the server 681 * @zero_point: The new value for the point beyond which the server has no data 682 * 683 * Set both remote_i_size and zero_point safely without the potential for 684 * tearing on 32-bit arches. 685 * 686 * Context: The caller must hold inode->i_lock. 687 * 688 * NOTE: unlike netfs_read_zero_point(), netfs_write_zero_point() does need 689 * locking around it (normally i_rwsem), otherwise on 32bit/SMP an update of 690 * i_size_seqcount can be lost, resulting in subsequent read calls spinning 691 * forever. 692 */ 693 static inline void netfs_write_sizes(struct inode *inode, 694 unsigned long long i_size, 695 unsigned long long remote_i_size, 696 unsigned long long zero_point) 697 { 698 struct netfs_inode *ictx = netfs_inode(inode); 699 700 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 701 write_seqcount_begin(&inode->i_size_seqcount); 702 inode->i_size = i_size; 703 ictx->_remote_i_size = remote_i_size; 704 ictx->_zero_point = zero_point; 705 write_seqcount_end(&inode->i_size_seqcount); 706 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 707 preempt_disable(); 708 inode->i_size = i_size; 709 ictx->_remote_i_size = remote_i_size; 710 ictx->_zero_point = zero_point; 711 preempt_enable(); 712 #else 713 /* 714 * Pairs with smp_load_acquire() in i_size_read(), 715 * netfs_read_remote_i_size() and netfs_read_zero_point() to ensure 716 * changes related to inode size (such as page contents) are visible 717 * before we see the changed inode size. 718 */ 719 smp_store_release(&inode->i_size, i_size); 720 smp_store_release(&ictx->_remote_i_size, remote_i_size); 721 smp_store_release(&ictx->_zero_point, zero_point); 722 #endif 723 } 724 725 /** 726 * netfs_inode_init - Initialise a netfslib inode context 727 * @ctx: The netfs inode to initialise 728 * @ops: The netfs's operations list 729 * @use_zero_point: True to use the zero_point read optimisation 730 * 731 * Initialise the netfs library context struct. This is expected to follow on 732 * directly from the VFS inode struct. 733 */ 734 static inline void netfs_inode_init(struct netfs_inode *ctx, 735 const struct netfs_request_ops *ops, 736 bool use_zero_point) 737 { 738 ctx->ops = ops; 739 ctx->_remote_i_size = i_size_read(&ctx->inode); 740 ctx->_zero_point = LLONG_MAX; 741 ctx->flags = 0; 742 atomic_set(&ctx->io_count, 0); 743 #if IS_ENABLED(CONFIG_FSCACHE) 744 ctx->cache = NULL; 745 #endif 746 mutex_init(&ctx->wb_lock); 747 /* ->releasepage() drives zero_point */ 748 if (use_zero_point) { 749 ctx->_zero_point = ctx->_remote_i_size; 750 mapping_set_release_always(ctx->inode.i_mapping); 751 } 752 } 753 754 /** 755 * netfs_resize_file - Note that a file got resized 756 * @ctx: The netfs inode being resized 757 * @new_i_size: The new file size 758 * @changed_on_server: The change was applied to the server 759 * 760 * Inform the netfs lib that a file got resized so that it can adjust its state. 761 */ 762 static inline void netfs_resize_file(struct netfs_inode *ictx, 763 unsigned long long new_i_size, 764 bool changed_on_server) 765 { 766 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 767 struct inode *inode = &ictx->inode; 768 769 preempt_disable(); 770 write_seqcount_begin(&inode->i_size_seqcount); 771 if (changed_on_server) 772 ictx->_remote_i_size = new_i_size; 773 if (new_i_size < ictx->_zero_point) 774 ictx->_zero_point = new_i_size; 775 write_seqcount_end(&inode->i_size_seqcount); 776 preempt_enable(); 777 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 778 preempt_disable(); 779 if (changed_on_server) 780 ictx->_remote_i_size = new_i_size; 781 if (new_i_size < ictx->_zero_point) 782 ictx->_zero_point = new_i_size; 783 preempt_enable(); 784 #else 785 /* 786 * Pairs with smp_load_acquire() in netfs_read_remote_i_size and 787 * netfs_read_zero_point() to ensure changes related to inode size 788 * (such as page contents) are visible before we see the changed inode 789 * size. 790 */ 791 if (changed_on_server) 792 smp_store_release(&ictx->_remote_i_size, new_i_size); 793 if (new_i_size < ictx->_zero_point) 794 smp_store_release(&ictx->_zero_point, new_i_size); 795 #endif 796 } 797 798 /** 799 * netfs_i_cookie - Get the cache cookie from the inode 800 * @ctx: The netfs inode to query 801 * 802 * Get the caching cookie (if enabled) from the network filesystem's inode. 803 */ 804 static inline struct fscache_cookie *netfs_i_cookie(struct netfs_inode *ctx) 805 { 806 #if IS_ENABLED(CONFIG_FSCACHE) 807 return ctx->cache; 808 #else 809 return NULL; 810 #endif 811 } 812 813 /** 814 * netfs_wait_for_outstanding_io - Wait for outstanding I/O to complete 815 * @inode: The netfs inode to wait on 816 * 817 * Wait for outstanding I/O requests of any type to complete. This is intended 818 * to be called from inode eviction routines. This makes sure that any 819 * resources held by those requests are cleaned up before we let the inode get 820 * cleaned up. 821 */ 822 static inline void netfs_wait_for_outstanding_io(struct inode *inode) 823 { 824 struct netfs_inode *ictx = netfs_inode(inode); 825 826 wait_var_event(&ictx->io_count, atomic_read(&ictx->io_count) == 0); 827 } 828 829 #endif /* _LINUX_NETFS_H */ 830