1.. SPDX-License-Identifier: GPL-2.0 2 3================================= 4Network Filesystem Helper Library 5================================= 6 7.. Contents: 8 9 - Overview. 10 - Per-inode context. 11 - Inode context helper functions. 12 - Buffered read helpers. 13 - Read helper functions. 14 - Read helper structures. 15 - Read helper operations. 16 - Read helper procedure. 17 - Read helper cache API. 18 19 20Overview 21======== 22 23The network filesystem helper library is a set of functions designed to aid a 24network filesystem in implementing VM/VFS operations. For the moment, that 25just includes turning various VM buffered read operations into requests to read 26from the server. The helper library, however, can also interpose other 27services, such as local caching or local data encryption. 28 29Note that the library module doesn't link against local caching directly, so 30access must be provided by the netfs. 31 32 33Per-Inode Context 34================= 35 36The network filesystem helper library needs a place to store a bit of state for 37its use on each netfs inode it is helping to manage. To this end, a context 38structure is defined:: 39 40 struct netfs_inode { 41 struct inode inode; 42 const struct netfs_request_ops *ops; 43 struct fscache_cookie *cache; 44 }; 45 46A network filesystem that wants to use netfs lib must place one of these in its 47inode wrapper struct instead of the VFS ``struct inode``. This can be done in 48a way similar to the following:: 49 50 struct my_inode { 51 struct netfs_inode netfs; /* Netfslib context and vfs inode */ 52 ... 53 }; 54 55This allows netfslib to find its state by using ``container_of()`` from the 56inode pointer, thereby allowing the netfslib helper functions to be pointed to 57directly by the VFS/VM operation tables. 58 59The structure contains the following fields: 60 61 * ``inode`` 62 63 The VFS inode structure. 64 65 * ``ops`` 66 67 The set of operations provided by the network filesystem to netfslib. 68 69 * ``cache`` 70 71 Local caching cookie, or NULL if no caching is enabled. This field does not 72 exist if fscache is disabled. 73 74 75Inode Context Helper Functions 76------------------------------ 77 78To help deal with the per-inode context, a number helper functions are 79provided. Firstly, a function to perform basic initialisation on a context and 80set the operations table pointer:: 81 82 void netfs_inode_init(struct netfs_inode *ctx, 83 const struct netfs_request_ops *ops); 84 85then a function to cast from the VFS inode structure to the netfs context:: 86 87 struct netfs_inode *netfs_node(struct inode *inode); 88 89and finally, a function to get the cache cookie pointer from the context 90attached to an inode (or NULL if fscache is disabled):: 91 92 struct fscache_cookie *netfs_i_cookie(struct netfs_inode *ctx); 93 94 95Buffered Read Helpers 96===================== 97 98The library provides a set of read helpers that handle the ->read_folio(), 99->readahead() and much of the ->write_begin() VM operations and translate them 100into a common call framework. 101 102The following services are provided: 103 104 * Handle folios that span multiple pages. 105 106 * Insulate the netfs from VM interface changes. 107 108 * Allow the netfs to arbitrarily split reads up into pieces, even ones that 109 don't match folio sizes or folio alignments and that may cross folios. 110 111 * Allow the netfs to expand a readahead request in both directions to meet its 112 needs. 113 114 * Allow the netfs to partially fulfil a read, which will then be resubmitted. 115 116 * Handle local caching, allowing cached data and server-read data to be 117 interleaved for a single request. 118 119 * Handle clearing of bufferage that aren't on the server. 120 121 * Handle retrying of reads that failed, switching reads from the cache to the 122 server as necessary. 123 124 * In the future, this is a place that other services can be performed, such as 125 local encryption of data to be stored remotely or in the cache. 126 127From the network filesystem, the helpers require a table of operations. This 128includes a mandatory method to issue a read operation along with a number of 129optional methods. 130 131 132Read Helper Functions 133--------------------- 134 135Three read helpers are provided:: 136 137 void netfs_readahead(struct readahead_control *ractl); 138 int netfs_read_folio(struct file *file, 139 struct folio *folio); 140 int netfs_write_begin(struct netfs_inode *ctx, 141 struct file *file, 142 struct address_space *mapping, 143 loff_t pos, 144 unsigned int len, 145 struct folio **_folio, 146 void **_fsdata); 147 148Each corresponds to a VM address space operation. These operations use the 149state in the per-inode context. 150 151For ->readahead() and ->read_folio(), the network filesystem just point directly 152at the corresponding read helper; whereas for ->write_begin(), it may be a 153little more complicated as the network filesystem might want to flush 154conflicting writes or track dirty data and needs to put the acquired folio if 155an error occurs after calling the helper. 156 157The helpers manage the read request, calling back into the network filesystem 158through the supplied table of operations. Waits will be performed as 159necessary before returning for helpers that are meant to be synchronous. 160 161If an error occurs, the ->free_request() will be called to clean up the 162netfs_io_request struct allocated. If some parts of the request are in 163progress when an error occurs, the request will get partially completed if 164sufficient data is read. 165 166Additionally, there is:: 167 168 * void netfs_subreq_terminated(struct netfs_io_subrequest *subreq, 169 ssize_t transferred_or_error, 170 bool was_async); 171 172which should be called to complete a read subrequest. This is given the number 173of bytes transferred or a negative error code, plus a flag indicating whether 174the operation was asynchronous (ie. whether the follow-on processing can be 175done in the current context, given this may involve sleeping). 176 177 178Read Helper Structures 179---------------------- 180 181The read helpers make use of a couple of structures to maintain the state of 182the read. The first is a structure that manages a read request as a whole:: 183 184 struct netfs_io_request { 185 struct inode *inode; 186 struct address_space *mapping; 187 struct netfs_cache_resources cache_resources; 188 void *netfs_priv; 189 loff_t start; 190 size_t len; 191 loff_t i_size; 192 const struct netfs_request_ops *netfs_ops; 193 unsigned int debug_id; 194 ... 195 }; 196 197The above fields are the ones the netfs can use. They are: 198 199 * ``inode`` 200 * ``mapping`` 201 202 The inode and the address space of the file being read from. The mapping 203 may or may not point to inode->i_data. 204 205 * ``cache_resources`` 206 207 Resources for the local cache to use, if present. 208 209 * ``netfs_priv`` 210 211 The network filesystem's private data. The value for this can be passed in 212 to the helper functions or set during the request. 213 214 * ``start`` 215 * ``len`` 216 217 The file position of the start of the read request and the length. These 218 may be altered by the ->expand_readahead() op. 219 220 * ``i_size`` 221 222 The size of the file at the start of the request. 223 224 * ``netfs_ops`` 225 226 A pointer to the operation table. The value for this is passed into the 227 helper functions. 228 229 * ``debug_id`` 230 231 A number allocated to this operation that can be displayed in trace lines 232 for reference. 233 234 235The second structure is used to manage individual slices of the overall read 236request:: 237 238 struct netfs_io_subrequest { 239 struct netfs_io_request *rreq; 240 loff_t start; 241 size_t len; 242 size_t transferred; 243 unsigned long flags; 244 unsigned short debug_index; 245 ... 246 }; 247 248Each subrequest is expected to access a single source, though the helpers will 249handle falling back from one source type to another. The members are: 250 251 * ``rreq`` 252 253 A pointer to the read request. 254 255 * ``start`` 256 * ``len`` 257 258 The file position of the start of this slice of the read request and the 259 length. 260 261 * ``transferred`` 262 263 The amount of data transferred so far of the length of this slice. The 264 network filesystem or cache should start the operation this far into the 265 slice. If a short read occurs, the helpers will call again, having updated 266 this to reflect the amount read so far. 267 268 * ``flags`` 269 270 Flags pertaining to the read. There are two of interest to the filesystem 271 or cache: 272 273 * ``NETFS_SREQ_CLEAR_TAIL`` 274 275 This can be set to indicate that the remainder of the slice, from 276 transferred to len, should be cleared. 277 278 * ``NETFS_SREQ_SEEK_DATA_READ`` 279 280 This is a hint to the cache that it might want to try skipping ahead to 281 the next data (ie. using SEEK_DATA). 282 283 * ``debug_index`` 284 285 A number allocated to this slice that can be displayed in trace lines for 286 reference. 287 288 289Read Helper Operations 290---------------------- 291 292The network filesystem must provide the read helpers with a table of operations 293through which it can issue requests and negotiate:: 294 295 struct netfs_request_ops { 296 void (*init_request)(struct netfs_io_request *rreq, struct file *file); 297 void (*free_request)(struct netfs_io_request *rreq); 298 void (*expand_readahead)(struct netfs_io_request *rreq); 299 bool (*clamp_length)(struct netfs_io_subrequest *subreq); 300 void (*issue_read)(struct netfs_io_subrequest *subreq); 301 bool (*is_still_valid)(struct netfs_io_request *rreq); 302 int (*check_write_begin)(struct file *file, loff_t pos, unsigned len, 303 struct folio **foliop, void **_fsdata); 304 void (*done)(struct netfs_io_request *rreq); 305 }; 306 307The operations are as follows: 308 309 * ``init_request()`` 310 311 [Optional] This is called to initialise the request structure. It is given 312 the file for reference. 313 314 * ``free_request()`` 315 316 [Optional] This is called as the request is being deallocated so that the 317 filesystem can clean up any state it has attached there. 318 319 * ``expand_readahead()`` 320 321 [Optional] This is called to allow the filesystem to expand the size of a 322 readahead read request. The filesystem gets to expand the request in both 323 directions, though it's not permitted to reduce it as the numbers may 324 represent an allocation already made. If local caching is enabled, it gets 325 to expand the request first. 326 327 Expansion is communicated by changing ->start and ->len in the request 328 structure. Note that if any change is made, ->len must be increased by at 329 least as much as ->start is reduced. 330 331 * ``clamp_length()`` 332 333 [Optional] This is called to allow the filesystem to reduce the size of a 334 subrequest. The filesystem can use this, for example, to chop up a request 335 that has to be split across multiple servers or to put multiple reads in 336 flight. 337 338 This should return 0 on success and an error code on error. 339 340 * ``issue_read()`` 341 342 [Required] The helpers use this to dispatch a subrequest to the server for 343 reading. In the subrequest, ->start, ->len and ->transferred indicate what 344 data should be read from the server. 345 346 There is no return value; the netfs_subreq_terminated() function should be 347 called to indicate whether or not the operation succeeded and how much data 348 it transferred. The filesystem also should not deal with setting folios 349 uptodate, unlocking them or dropping their refs - the helpers need to deal 350 with this as they have to coordinate with copying to the local cache. 351 352 Note that the helpers have the folios locked, but not pinned. It is 353 possible to use the ITER_XARRAY iov iterator to refer to the range of the 354 inode that is being operated upon without the need to allocate large bvec 355 tables. 356 357 * ``is_still_valid()`` 358 359 [Optional] This is called to find out if the data just read from the local 360 cache is still valid. It should return true if it is still valid and false 361 if not. If it's not still valid, it will be reread from the server. 362 363 * ``check_write_begin()`` 364 365 [Optional] This is called from the netfs_write_begin() helper once it has 366 allocated/grabbed the folio to be modified to allow the filesystem to flush 367 conflicting state before allowing it to be modified. 368 369 It may unlock and discard the folio it was given and set the caller's folio 370 pointer to NULL. It should return 0 if everything is now fine (``*foliop`` 371 left set) or the op should be retried (``*foliop`` cleared) and any other 372 error code to abort the operation. 373 374 * ``done`` 375 376 [Optional] This is called after the folios in the request have all been 377 unlocked (and marked uptodate if applicable). 378 379 380 381Read Helper Procedure 382--------------------- 383 384The read helpers work by the following general procedure: 385 386 * Set up the request. 387 388 * For readahead, allow the local cache and then the network filesystem to 389 propose expansions to the read request. This is then proposed to the VM. 390 If the VM cannot fully perform the expansion, a partially expanded read will 391 be performed, though this may not get written to the cache in its entirety. 392 393 * Loop around slicing chunks off of the request to form subrequests: 394 395 * If a local cache is present, it gets to do the slicing, otherwise the 396 helpers just try to generate maximal slices. 397 398 * The network filesystem gets to clamp the size of each slice if it is to be 399 the source. This allows rsize and chunking to be implemented. 400 401 * The helpers issue a read from the cache or a read from the server or just 402 clears the slice as appropriate. 403 404 * The next slice begins at the end of the last one. 405 406 * As slices finish being read, they terminate. 407 408 * When all the subrequests have terminated, the subrequests are assessed and 409 any that are short or have failed are reissued: 410 411 * Failed cache requests are issued against the server instead. 412 413 * Failed server requests just fail. 414 415 * Short reads against either source will be reissued against that source 416 provided they have transferred some more data: 417 418 * The cache may need to skip holes that it can't do DIO from. 419 420 * If NETFS_SREQ_CLEAR_TAIL was set, a short read will be cleared to the 421 end of the slice instead of reissuing. 422 423 * Once the data is read, the folios that have been fully read/cleared: 424 425 * Will be marked uptodate. 426 427 * If a cache is present, will be marked with PG_fscache. 428 429 * Unlocked 430 431 * Any folios that need writing to the cache will then have DIO writes issued. 432 433 * Synchronous operations will wait for reading to be complete. 434 435 * Writes to the cache will proceed asynchronously and the folios will have the 436 PG_fscache mark removed when that completes. 437 438 * The request structures will be cleaned up when everything has completed. 439 440 441Read Helper Cache API 442--------------------- 443 444When implementing a local cache to be used by the read helpers, two things are 445required: some way for the network filesystem to initialise the caching for a 446read request and a table of operations for the helpers to call. 447 448To begin a cache operation on an fscache object, the following function is 449called:: 450 451 int fscache_begin_read_operation(struct netfs_io_request *rreq, 452 struct fscache_cookie *cookie); 453 454passing in the request pointer and the cookie corresponding to the file. This 455fills in the cache resources mentioned below. 456 457The netfs_io_request object contains a place for the cache to hang its 458state:: 459 460 struct netfs_cache_resources { 461 const struct netfs_cache_ops *ops; 462 void *cache_priv; 463 void *cache_priv2; 464 }; 465 466This contains an operations table pointer and two private pointers. The 467operation table looks like the following:: 468 469 struct netfs_cache_ops { 470 void (*end_operation)(struct netfs_cache_resources *cres); 471 472 void (*expand_readahead)(struct netfs_cache_resources *cres, 473 loff_t *_start, size_t *_len, loff_t i_size); 474 475 enum netfs_io_source (*prepare_read)(struct netfs_io_subrequest *subreq, 476 loff_t i_size); 477 478 int (*read)(struct netfs_cache_resources *cres, 479 loff_t start_pos, 480 struct iov_iter *iter, 481 bool seek_data, 482 netfs_io_terminated_t term_func, 483 void *term_func_priv); 484 485 int (*prepare_write)(struct netfs_cache_resources *cres, 486 loff_t *_start, size_t *_len, loff_t i_size, 487 bool no_space_allocated_yet); 488 489 int (*write)(struct netfs_cache_resources *cres, 490 loff_t start_pos, 491 struct iov_iter *iter, 492 netfs_io_terminated_t term_func, 493 void *term_func_priv); 494 495 int (*query_occupancy)(struct netfs_cache_resources *cres, 496 loff_t start, size_t len, size_t granularity, 497 loff_t *_data_start, size_t *_data_len); 498 }; 499 500With a termination handler function pointer:: 501 502 typedef void (*netfs_io_terminated_t)(void *priv, 503 ssize_t transferred_or_error, 504 bool was_async); 505 506The methods defined in the table are: 507 508 * ``end_operation()`` 509 510 [Required] Called to clean up the resources at the end of the read request. 511 512 * ``expand_readahead()`` 513 514 [Optional] Called at the beginning of a netfs_readahead() operation to allow 515 the cache to expand a request in either direction. This allows the cache to 516 size the request appropriately for the cache granularity. 517 518 The function is passed poiners to the start and length in its parameters, 519 plus the size of the file for reference, and adjusts the start and length 520 appropriately. It should return one of: 521 522 * ``NETFS_FILL_WITH_ZEROES`` 523 * ``NETFS_DOWNLOAD_FROM_SERVER`` 524 * ``NETFS_READ_FROM_CACHE`` 525 * ``NETFS_INVALID_READ`` 526 527 to indicate whether the slice should just be cleared or whether it should be 528 downloaded from the server or read from the cache - or whether slicing 529 should be given up at the current point. 530 531 * ``prepare_read()`` 532 533 [Required] Called to configure the next slice of a request. ->start and 534 ->len in the subrequest indicate where and how big the next slice can be; 535 the cache gets to reduce the length to match its granularity requirements. 536 537 * ``read()`` 538 539 [Required] Called to read from the cache. The start file offset is given 540 along with an iterator to read to, which gives the length also. It can be 541 given a hint requesting that it seek forward from that start position for 542 data. 543 544 Also provided is a pointer to a termination handler function and private 545 data to pass to that function. The termination function should be called 546 with the number of bytes transferred or an error code, plus a flag 547 indicating whether the termination is definitely happening in the caller's 548 context. 549 550 * ``prepare_write()`` 551 552 [Required] Called to prepare a write to the cache to take place. This 553 involves checking to see whether the cache has sufficient space to honour 554 the write. ``*_start`` and ``*_len`` indicate the region to be written; the 555 region can be shrunk or it can be expanded to a page boundary either way as 556 necessary to align for direct I/O. i_size holds the size of the object and 557 is provided for reference. no_space_allocated_yet is set to true if the 558 caller is certain that no data has been written to that region - for example 559 if it tried to do a read from there already. 560 561 * ``write()`` 562 563 [Required] Called to write to the cache. The start file offset is given 564 along with an iterator to write from, which gives the length also. 565 566 Also provided is a pointer to a termination handler function and private 567 data to pass to that function. The termination function should be called 568 with the number of bytes transferred or an error code, plus a flag 569 indicating whether the termination is definitely happening in the caller's 570 context. 571 572 * ``query_occupancy()`` 573 574 [Required] Called to find out where the next piece of data is within a 575 particular region of the cache. The start and length of the region to be 576 queried are passed in, along with the granularity to which the answer needs 577 to be aligned. The function passes back the start and length of the data, 578 if any, available within that region. Note that there may be a hole at the 579 front. 580 581 It returns 0 if some data was found, -ENODATA if there was no usable data 582 within the region or -ENOBUFS if there is no caching on this file. 583 584Note that these methods are passed a pointer to the cache resource structure, 585not the read request structure as they could be used in other situations where 586there isn't a read request structure as well, such as writing dirty data to the 587cache. 588 589 590API Function Reference 591====================== 592 593.. kernel-doc:: include/linux/netfs.h 594.. kernel-doc:: fs/netfs/buffered_read.c 595.. kernel-doc:: fs/netfs/io.c 596