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 ([email protected]) 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 22 enum netfs_sreq_ref_trace; 23 typedef struct mempool_s mempool_t; 24 25 /** 26 * folio_start_private_2 - Start an fscache write on a folio. [DEPRECATED] 27 * @folio: The folio. 28 * 29 * Call this function before writing a folio to a local cache. Starting a 30 * second write before the first one finishes is not allowed. 31 * 32 * Note that this should no longer be used. 33 */ 34 static inline void folio_start_private_2(struct folio *folio) 35 { 36 VM_BUG_ON_FOLIO(folio_test_private_2(folio), folio); 37 folio_get(folio); 38 folio_set_private_2(folio); 39 } 40 41 /* Marks used on xarray-based buffers */ 42 #define NETFS_BUF_PUT_MARK XA_MARK_0 /* - Page needs putting */ 43 #define NETFS_BUF_PAGECACHE_MARK XA_MARK_1 /* - Page needs wb/dirty flag wrangling */ 44 45 enum netfs_io_source { 46 NETFS_FILL_WITH_ZEROES, 47 NETFS_DOWNLOAD_FROM_SERVER, 48 NETFS_READ_FROM_CACHE, 49 NETFS_INVALID_READ, 50 NETFS_UPLOAD_TO_SERVER, 51 NETFS_WRITE_TO_CACHE, 52 NETFS_INVALID_WRITE, 53 } __mode(byte); 54 55 typedef void (*netfs_io_terminated_t)(void *priv, ssize_t transferred_or_error, 56 bool was_async); 57 58 /* 59 * Per-inode context. This wraps the VFS inode. 60 */ 61 struct netfs_inode { 62 struct inode inode; /* The VFS inode */ 63 const struct netfs_request_ops *ops; 64 #if IS_ENABLED(CONFIG_FSCACHE) 65 struct fscache_cookie *cache; 66 #endif 67 struct mutex wb_lock; /* Writeback serialisation */ 68 loff_t remote_i_size; /* Size of the remote file */ 69 loff_t zero_point; /* Size after which we assume there's no data 70 * on the server */ 71 unsigned long flags; 72 #define NETFS_ICTX_ODIRECT 0 /* The file has DIO in progress */ 73 #define NETFS_ICTX_UNBUFFERED 1 /* I/O should not use the pagecache */ 74 #define NETFS_ICTX_WRITETHROUGH 2 /* Write-through caching */ 75 #define NETFS_ICTX_USE_PGPRIV2 31 /* [DEPRECATED] Use PG_private_2 to mark 76 * write to cache on read */ 77 }; 78 79 /* 80 * A netfs group - for instance a ceph snap. This is marked on dirty pages and 81 * pages marked with a group must be flushed before they can be written under 82 * the domain of another group. 83 */ 84 struct netfs_group { 85 refcount_t ref; 86 void (*free)(struct netfs_group *netfs_group); 87 }; 88 89 /* 90 * Information about a dirty page (attached only if necessary). 91 * folio->private 92 */ 93 struct netfs_folio { 94 struct netfs_group *netfs_group; /* Filesystem's grouping marker (or NULL). */ 95 unsigned int dirty_offset; /* Write-streaming dirty data offset */ 96 unsigned int dirty_len; /* Write-streaming dirty data length */ 97 }; 98 #define NETFS_FOLIO_INFO 0x1UL /* OR'd with folio->private. */ 99 #define NETFS_FOLIO_COPY_TO_CACHE ((struct netfs_group *)0x356UL) /* Write to the cache only */ 100 101 static inline bool netfs_is_folio_info(const void *priv) 102 { 103 return (unsigned long)priv & NETFS_FOLIO_INFO; 104 } 105 106 static inline struct netfs_folio *__netfs_folio_info(const void *priv) 107 { 108 if (netfs_is_folio_info(priv)) 109 return (struct netfs_folio *)((unsigned long)priv & ~NETFS_FOLIO_INFO); 110 return NULL; 111 } 112 113 static inline struct netfs_folio *netfs_folio_info(struct folio *folio) 114 { 115 return __netfs_folio_info(folio_get_private(folio)); 116 } 117 118 static inline struct netfs_group *netfs_folio_group(struct folio *folio) 119 { 120 struct netfs_folio *finfo; 121 void *priv = folio_get_private(folio); 122 123 finfo = netfs_folio_info(folio); 124 if (finfo) 125 return finfo->netfs_group; 126 return priv; 127 } 128 129 /* 130 * Stream of I/O subrequests going to a particular destination, such as the 131 * server or the local cache. This is mainly intended for writing where we may 132 * have to write to multiple destinations concurrently. 133 */ 134 struct netfs_io_stream { 135 /* Submission tracking */ 136 struct netfs_io_subrequest *construct; /* Op being constructed */ 137 unsigned int submit_off; /* Folio offset we're submitting from */ 138 unsigned int submit_len; /* Amount of data left to submit */ 139 unsigned int submit_max_len; /* Amount I/O can be rounded up to */ 140 void (*prepare_write)(struct netfs_io_subrequest *subreq); 141 void (*issue_write)(struct netfs_io_subrequest *subreq); 142 /* Collection tracking */ 143 struct list_head subrequests; /* Contributory I/O operations */ 144 struct netfs_io_subrequest *front; /* Op being collected */ 145 unsigned long long collected_to; /* Position we've collected results to */ 146 size_t transferred; /* The amount transferred from this stream */ 147 enum netfs_io_source source; /* Where to read from/write to */ 148 unsigned short error; /* Aggregate error for the stream */ 149 unsigned char stream_nr; /* Index of stream in parent table */ 150 bool avail; /* T if stream is available */ 151 bool active; /* T if stream is active */ 152 bool need_retry; /* T if this stream needs retrying */ 153 bool failed; /* T if this stream failed */ 154 }; 155 156 /* 157 * Resources required to do operations on a cache. 158 */ 159 struct netfs_cache_resources { 160 const struct netfs_cache_ops *ops; 161 void *cache_priv; 162 void *cache_priv2; 163 unsigned int debug_id; /* Cookie debug ID */ 164 unsigned int inval_counter; /* object->inval_counter at begin_op */ 165 }; 166 167 /* 168 * Descriptor for a single component subrequest. Each operation represents an 169 * individual read/write from/to a server, a cache, a journal, etc.. 170 * 171 * The buffer iterator is persistent for the life of the subrequest struct and 172 * the pages it points to can be relied on to exist for the duration. 173 */ 174 struct netfs_io_subrequest { 175 struct netfs_io_request *rreq; /* Supervising I/O request */ 176 struct work_struct work; 177 struct list_head rreq_link; /* Link in rreq->subrequests */ 178 struct iov_iter io_iter; /* Iterator for this subrequest */ 179 unsigned long long start; /* Where to start the I/O */ 180 size_t max_len; /* Maximum size of the I/O */ 181 size_t len; /* Size of the I/O */ 182 size_t transferred; /* Amount of data transferred */ 183 refcount_t ref; 184 short error; /* 0 or error that occurred */ 185 unsigned short debug_index; /* Index in list (for debugging output) */ 186 unsigned int nr_segs; /* Number of segs in io_iter */ 187 unsigned int max_nr_segs; /* 0 or max number of segments in an iterator */ 188 enum netfs_io_source source; /* Where to read from/write to */ 189 unsigned char stream_nr; /* I/O stream this belongs to */ 190 unsigned long flags; 191 #define NETFS_SREQ_COPY_TO_CACHE 0 /* Set if should copy the data to the cache */ 192 #define NETFS_SREQ_CLEAR_TAIL 1 /* Set if the rest of the read should be cleared */ 193 #define NETFS_SREQ_SHORT_IO 2 /* Set if the I/O was short */ 194 #define NETFS_SREQ_SEEK_DATA_READ 3 /* Set if ->read() should SEEK_DATA first */ 195 #define NETFS_SREQ_NO_PROGRESS 4 /* Set if we didn't manage to read any data */ 196 #define NETFS_SREQ_ONDEMAND 5 /* Set if it's from on-demand read mode */ 197 #define NETFS_SREQ_BOUNDARY 6 /* Set if ends on hard boundary (eg. ceph object) */ 198 #define NETFS_SREQ_IN_PROGRESS 8 /* Unlocked when the subrequest completes */ 199 #define NETFS_SREQ_NEED_RETRY 9 /* Set if the filesystem requests a retry */ 200 #define NETFS_SREQ_RETRYING 10 /* Set if we're retrying */ 201 #define NETFS_SREQ_FAILED 11 /* Set if the subreq failed unretryably */ 202 }; 203 204 enum netfs_io_origin { 205 NETFS_READAHEAD, /* This read was triggered by readahead */ 206 NETFS_READPAGE, /* This read is a synchronous read */ 207 NETFS_READ_FOR_WRITE, /* This read is to prepare a write */ 208 NETFS_COPY_TO_CACHE, /* This write is to copy a read to the cache */ 209 NETFS_WRITEBACK, /* This 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_READ, /* This is a direct I/O read */ 213 NETFS_DIO_WRITE, /* This is a direct I/O write */ 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 work; 224 struct rcu_head rcu; 225 }; 226 struct inode *inode; /* The file being accessed */ 227 struct address_space *mapping; /* The mapping being accessed */ 228 struct kiocb *iocb; /* AIO completion vector */ 229 struct netfs_cache_resources cache_resources; 230 struct list_head proc_link; /* Link in netfs_iorequests */ 231 struct list_head subrequests; /* Contributory I/O operations */ 232 struct netfs_io_stream io_streams[2]; /* Streams of parallel I/O operations */ 233 #define NR_IO_STREAMS 2 //wreq->nr_io_streams 234 struct netfs_group *group; /* Writeback group being written back */ 235 struct iov_iter iter; /* Unencrypted-side iterator */ 236 struct iov_iter io_iter; /* I/O (Encrypted-side) iterator */ 237 void *netfs_priv; /* Private data for the netfs */ 238 void *netfs_priv2; /* Private data for the netfs */ 239 struct bio_vec *direct_bv; /* DIO buffer list (when handling iovec-iter) */ 240 unsigned int direct_bv_count; /* Number of elements in direct_bv[] */ 241 unsigned int debug_id; 242 unsigned int rsize; /* Maximum read size (0 for none) */ 243 unsigned int wsize; /* Maximum write size (0 for none) */ 244 atomic_t subreq_counter; /* Next subreq->debug_index */ 245 unsigned int nr_group_rel; /* Number of refs to release on ->group */ 246 spinlock_t lock; /* Lock for queuing subreqs */ 247 atomic_t nr_outstanding; /* Number of ops in progress */ 248 atomic_t nr_copy_ops; /* Number of copy-to-cache ops in progress */ 249 size_t upper_len; /* Length can be extended to here */ 250 unsigned long long submitted; /* Amount submitted for I/O so far */ 251 unsigned long long len; /* Length of the request */ 252 size_t transferred; /* Amount to be indicated as transferred */ 253 short error; /* 0 or error that occurred */ 254 enum netfs_io_origin origin; /* Origin of the request */ 255 bool direct_bv_unpin; /* T if direct_bv[] must be unpinned */ 256 unsigned long long i_size; /* Size of the file */ 257 unsigned long long start; /* Start position */ 258 atomic64_t issued_to; /* Write issuer folio cursor */ 259 unsigned long long contiguity; /* Tracking for gaps in the writeback sequence */ 260 unsigned long long collected_to; /* Point we've collected to */ 261 unsigned long long cleaned_to; /* Position we've cleaned folios to */ 262 pgoff_t no_unlock_folio; /* Don't unlock this folio after read */ 263 refcount_t ref; 264 unsigned long flags; 265 #define NETFS_RREQ_INCOMPLETE_IO 0 /* Some ioreqs terminated short or with error */ 266 #define NETFS_RREQ_COPY_TO_CACHE 1 /* Need to write to the cache */ 267 #define NETFS_RREQ_NO_UNLOCK_FOLIO 2 /* Don't unlock no_unlock_folio on completion */ 268 #define NETFS_RREQ_DONT_UNLOCK_FOLIOS 3 /* Don't unlock the folios on completion */ 269 #define NETFS_RREQ_FAILED 4 /* The request failed */ 270 #define NETFS_RREQ_IN_PROGRESS 5 /* Unlocked when the request completes */ 271 #define NETFS_RREQ_WRITE_TO_CACHE 7 /* Need to write to the cache */ 272 #define NETFS_RREQ_UPLOAD_TO_SERVER 8 /* Need to write to the server */ 273 #define NETFS_RREQ_NONBLOCK 9 /* Don't block if possible (O_NONBLOCK) */ 274 #define NETFS_RREQ_BLOCKED 10 /* We blocked */ 275 #define NETFS_RREQ_PAUSE 11 /* Pause subrequest generation */ 276 #define NETFS_RREQ_USE_IO_ITER 12 /* Use ->io_iter rather than ->i_pages */ 277 #define NETFS_RREQ_ALL_QUEUED 13 /* All subreqs are now queued */ 278 #define NETFS_RREQ_USE_PGPRIV2 31 /* [DEPRECATED] Use PG_private_2 to mark 279 * write to cache on read */ 280 const struct netfs_request_ops *netfs_ops; 281 void (*cleanup)(struct netfs_io_request *req); 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 bool (*clamp_length)(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_CLEAR, 321 NETFS_READ_HOLE_FAIL, 322 }; 323 324 /* 325 * Table of operations for access to a cache. 326 */ 327 struct netfs_cache_ops { 328 /* End an operation */ 329 void (*end_operation)(struct netfs_cache_resources *cres); 330 331 /* Read data from the cache */ 332 int (*read)(struct netfs_cache_resources *cres, 333 loff_t start_pos, 334 struct iov_iter *iter, 335 enum netfs_read_from_hole read_hole, 336 netfs_io_terminated_t term_func, 337 void *term_func_priv); 338 339 /* Write data to the cache */ 340 int (*write)(struct netfs_cache_resources *cres, 341 loff_t start_pos, 342 struct iov_iter *iter, 343 netfs_io_terminated_t term_func, 344 void *term_func_priv); 345 346 /* Write data to the cache from a netfs subrequest. */ 347 void (*issue_write)(struct netfs_io_subrequest *subreq); 348 349 /* Expand readahead request */ 350 void (*expand_readahead)(struct netfs_cache_resources *cres, 351 unsigned long long *_start, 352 unsigned long long *_len, 353 unsigned long long i_size); 354 355 /* Prepare a read operation, shortening it to a cached/uncached 356 * boundary as appropriate. 357 */ 358 enum netfs_io_source (*prepare_read)(struct netfs_io_subrequest *subreq, 359 unsigned long long i_size); 360 361 /* Prepare a write subrequest, working out if we're allowed to do it 362 * and finding out the maximum amount of data to gather before 363 * attempting to submit. If we're not permitted to do it, the 364 * subrequest should be marked failed. 365 */ 366 void (*prepare_write_subreq)(struct netfs_io_subrequest *subreq); 367 368 /* Prepare a write operation, working out what part of the write we can 369 * actually do. 370 */ 371 int (*prepare_write)(struct netfs_cache_resources *cres, 372 loff_t *_start, size_t *_len, size_t upper_len, 373 loff_t i_size, bool no_space_allocated_yet); 374 375 /* Prepare an on-demand read operation, shortening it to a cached/uncached 376 * boundary as appropriate. 377 */ 378 enum netfs_io_source (*prepare_ondemand_read)(struct netfs_cache_resources *cres, 379 loff_t start, size_t *_len, 380 loff_t i_size, 381 unsigned long *_flags, ino_t ino); 382 383 /* Query the occupancy of the cache in a region, returning where the 384 * next chunk of data starts and how long it is. 385 */ 386 int (*query_occupancy)(struct netfs_cache_resources *cres, 387 loff_t start, size_t len, size_t granularity, 388 loff_t *_data_start, size_t *_data_len); 389 }; 390 391 /* High-level read API. */ 392 ssize_t netfs_unbuffered_read_iter_locked(struct kiocb *iocb, struct iov_iter *iter); 393 ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter); 394 ssize_t netfs_buffered_read_iter(struct kiocb *iocb, struct iov_iter *iter); 395 ssize_t netfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter); 396 397 /* High-level write API */ 398 ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter, 399 struct netfs_group *netfs_group); 400 ssize_t netfs_buffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *from, 401 struct netfs_group *netfs_group); 402 ssize_t netfs_unbuffered_write_iter(struct kiocb *iocb, struct iov_iter *from); 403 ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *iter, 404 struct netfs_group *netfs_group); 405 ssize_t netfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from); 406 407 /* Address operations API */ 408 struct readahead_control; 409 void netfs_readahead(struct readahead_control *); 410 int netfs_read_folio(struct file *, struct folio *); 411 int netfs_write_begin(struct netfs_inode *, struct file *, 412 struct address_space *, loff_t pos, unsigned int len, 413 struct folio **, void **fsdata); 414 int netfs_writepages(struct address_space *mapping, 415 struct writeback_control *wbc); 416 bool netfs_dirty_folio(struct address_space *mapping, struct folio *folio); 417 int netfs_unpin_writeback(struct inode *inode, struct writeback_control *wbc); 418 void netfs_clear_inode_writeback(struct inode *inode, const void *aux); 419 void netfs_invalidate_folio(struct folio *folio, size_t offset, size_t length); 420 bool netfs_release_folio(struct folio *folio, gfp_t gfp); 421 422 /* VMA operations API. */ 423 vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group); 424 425 /* (Sub)request management API. */ 426 void netfs_subreq_terminated(struct netfs_io_subrequest *, ssize_t, bool); 427 void netfs_get_subrequest(struct netfs_io_subrequest *subreq, 428 enum netfs_sreq_ref_trace what); 429 void netfs_put_subrequest(struct netfs_io_subrequest *subreq, 430 bool was_async, enum netfs_sreq_ref_trace what); 431 ssize_t netfs_extract_user_iter(struct iov_iter *orig, size_t orig_len, 432 struct iov_iter *new, 433 iov_iter_extraction_t extraction_flags); 434 size_t netfs_limit_iter(const struct iov_iter *iter, size_t start_offset, 435 size_t max_size, size_t max_segs); 436 void netfs_prepare_write_failed(struct netfs_io_subrequest *subreq); 437 void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error, 438 bool was_async); 439 void netfs_queue_write_request(struct netfs_io_subrequest *subreq); 440 441 int netfs_start_io_read(struct inode *inode); 442 void netfs_end_io_read(struct inode *inode); 443 int netfs_start_io_write(struct inode *inode); 444 void netfs_end_io_write(struct inode *inode); 445 int netfs_start_io_direct(struct inode *inode); 446 void netfs_end_io_direct(struct inode *inode); 447 448 /** 449 * netfs_inode - Get the netfs inode context from the inode 450 * @inode: The inode to query 451 * 452 * Get the netfs lib inode context from the network filesystem's inode. The 453 * context struct is expected to directly follow on from the VFS inode struct. 454 */ 455 static inline struct netfs_inode *netfs_inode(struct inode *inode) 456 { 457 return container_of(inode, struct netfs_inode, inode); 458 } 459 460 /** 461 * netfs_inode_init - Initialise a netfslib inode context 462 * @ctx: The netfs inode to initialise 463 * @ops: The netfs's operations list 464 * @use_zero_point: True to use the zero_point read optimisation 465 * 466 * Initialise the netfs library context struct. This is expected to follow on 467 * directly from the VFS inode struct. 468 */ 469 static inline void netfs_inode_init(struct netfs_inode *ctx, 470 const struct netfs_request_ops *ops, 471 bool use_zero_point) 472 { 473 ctx->ops = ops; 474 ctx->remote_i_size = i_size_read(&ctx->inode); 475 ctx->zero_point = LLONG_MAX; 476 ctx->flags = 0; 477 #if IS_ENABLED(CONFIG_FSCACHE) 478 ctx->cache = NULL; 479 #endif 480 mutex_init(&ctx->wb_lock); 481 /* ->releasepage() drives zero_point */ 482 if (use_zero_point) { 483 ctx->zero_point = ctx->remote_i_size; 484 mapping_set_release_always(ctx->inode.i_mapping); 485 } 486 } 487 488 /** 489 * netfs_resize_file - Note that a file got resized 490 * @ctx: The netfs inode being resized 491 * @new_i_size: The new file size 492 * @changed_on_server: The change was applied to the server 493 * 494 * Inform the netfs lib that a file got resized so that it can adjust its state. 495 */ 496 static inline void netfs_resize_file(struct netfs_inode *ctx, loff_t new_i_size, 497 bool changed_on_server) 498 { 499 if (changed_on_server) 500 ctx->remote_i_size = new_i_size; 501 if (new_i_size < ctx->zero_point) 502 ctx->zero_point = new_i_size; 503 } 504 505 /** 506 * netfs_i_cookie - Get the cache cookie from the inode 507 * @ctx: The netfs inode to query 508 * 509 * Get the caching cookie (if enabled) from the network filesystem's inode. 510 */ 511 static inline struct fscache_cookie *netfs_i_cookie(struct netfs_inode *ctx) 512 { 513 #if IS_ENABLED(CONFIG_FSCACHE) 514 return ctx->cache; 515 #else 516 return NULL; 517 #endif 518 } 519 520 #endif /* _LINUX_NETFS_H */ 521