1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/9p/vfs_file.c 4 * 5 * This file contians vfs file ops for 9P2000. 6 * 7 * Copyright (C) 2004 by Eric Van Hensbergen <ericvh@gmail.com> 8 * Copyright (C) 2002 by Ron Minnich <rminnich@lanl.gov> 9 */ 10 11 #include <linux/module.h> 12 #include <linux/errno.h> 13 #include <linux/fs.h> 14 #include <linux/sched.h> 15 #include <linux/file.h> 16 #include <linux/stat.h> 17 #include <linux/string.h> 18 #include <linux/inet.h> 19 #include <linux/list.h> 20 #include <linux/pagemap.h> 21 #include <linux/utsname.h> 22 #include <linux/uaccess.h> 23 #include <linux/idr.h> 24 #include <linux/uio.h> 25 #include <linux/slab.h> 26 #include <net/9p/9p.h> 27 #include <net/9p/client.h> 28 29 #include "v9fs.h" 30 #include "v9fs_vfs.h" 31 #include "fid.h" 32 #include "cache.h" 33 34 static const struct vm_operations_struct v9fs_file_vm_ops; 35 static const struct vm_operations_struct v9fs_mmap_file_vm_ops; 36 37 /** 38 * v9fs_file_open - open a file (or directory) 39 * @inode: inode to be opened 40 * @file: file being opened 41 * 42 */ 43 44 int v9fs_file_open(struct inode *inode, struct file *file) 45 { 46 int err; 47 struct v9fs_inode *v9inode; 48 struct v9fs_session_info *v9ses; 49 struct p9_fid *fid; 50 int omode; 51 52 p9_debug(P9_DEBUG_VFS, "inode: %p file: %p\n", inode, file); 53 v9inode = V9FS_I(inode); 54 v9ses = v9fs_inode2v9ses(inode); 55 if (v9fs_proto_dotl(v9ses)) 56 omode = v9fs_open_to_dotl_flags(file->f_flags); 57 else 58 omode = v9fs_uflags2omode(file->f_flags, 59 v9fs_proto_dotu(v9ses)); 60 fid = file->private_data; 61 if (!fid) { 62 fid = v9fs_fid_clone(file_dentry(file)); 63 if (IS_ERR(fid)) 64 return PTR_ERR(fid); 65 66 err = p9_client_open(fid, omode); 67 if (err < 0) { 68 p9_client_clunk(fid); 69 return err; 70 } 71 if ((file->f_flags & O_APPEND) && 72 (!v9fs_proto_dotu(v9ses) && !v9fs_proto_dotl(v9ses))) 73 generic_file_llseek(file, 0, SEEK_END); 74 } 75 76 file->private_data = fid; 77 mutex_lock(&v9inode->v_mutex); 78 if ((v9ses->cache == CACHE_LOOSE || v9ses->cache == CACHE_FSCACHE) && 79 !v9inode->writeback_fid && 80 ((file->f_flags & O_ACCMODE) != O_RDONLY)) { 81 /* 82 * clone a fid and add it to writeback_fid 83 * we do it during open time instead of 84 * page dirty time via write_begin/page_mkwrite 85 * because we want write after unlink usecase 86 * to work. 87 */ 88 fid = v9fs_writeback_fid(file_dentry(file)); 89 if (IS_ERR(fid)) { 90 err = PTR_ERR(fid); 91 mutex_unlock(&v9inode->v_mutex); 92 goto out_error; 93 } 94 v9inode->writeback_fid = (void *) fid; 95 } 96 mutex_unlock(&v9inode->v_mutex); 97 if (v9ses->cache == CACHE_LOOSE || v9ses->cache == CACHE_FSCACHE) 98 v9fs_cache_inode_set_cookie(inode, file); 99 return 0; 100 out_error: 101 p9_client_clunk(file->private_data); 102 file->private_data = NULL; 103 return err; 104 } 105 106 /** 107 * v9fs_file_lock - lock a file (or directory) 108 * @filp: file to be locked 109 * @cmd: lock command 110 * @fl: file lock structure 111 * 112 * Bugs: this looks like a local only lock, we should extend into 9P 113 * by using open exclusive 114 */ 115 116 static int v9fs_file_lock(struct file *filp, int cmd, struct file_lock *fl) 117 { 118 int res = 0; 119 struct inode *inode = file_inode(filp); 120 121 p9_debug(P9_DEBUG_VFS, "filp: %p lock: %p\n", filp, fl); 122 123 /* No mandatory locks */ 124 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) 125 return -ENOLCK; 126 127 if ((IS_SETLK(cmd) || IS_SETLKW(cmd)) && fl->fl_type != F_UNLCK) { 128 filemap_write_and_wait(inode->i_mapping); 129 invalidate_mapping_pages(&inode->i_data, 0, -1); 130 } 131 132 return res; 133 } 134 135 static int v9fs_file_do_lock(struct file *filp, int cmd, struct file_lock *fl) 136 { 137 struct p9_flock flock; 138 struct p9_fid *fid; 139 uint8_t status = P9_LOCK_ERROR; 140 int res = 0; 141 unsigned char fl_type; 142 struct v9fs_session_info *v9ses; 143 144 fid = filp->private_data; 145 BUG_ON(fid == NULL); 146 147 if ((fl->fl_flags & FL_POSIX) != FL_POSIX) 148 BUG(); 149 150 res = locks_lock_file_wait(filp, fl); 151 if (res < 0) 152 goto out; 153 154 /* convert posix lock to p9 tlock args */ 155 memset(&flock, 0, sizeof(flock)); 156 /* map the lock type */ 157 switch (fl->fl_type) { 158 case F_RDLCK: 159 flock.type = P9_LOCK_TYPE_RDLCK; 160 break; 161 case F_WRLCK: 162 flock.type = P9_LOCK_TYPE_WRLCK; 163 break; 164 case F_UNLCK: 165 flock.type = P9_LOCK_TYPE_UNLCK; 166 break; 167 } 168 flock.start = fl->fl_start; 169 if (fl->fl_end == OFFSET_MAX) 170 flock.length = 0; 171 else 172 flock.length = fl->fl_end - fl->fl_start + 1; 173 flock.proc_id = fl->fl_pid; 174 flock.client_id = fid->clnt->name; 175 if (IS_SETLKW(cmd)) 176 flock.flags = P9_LOCK_FLAGS_BLOCK; 177 178 v9ses = v9fs_inode2v9ses(file_inode(filp)); 179 180 /* 181 * if its a blocked request and we get P9_LOCK_BLOCKED as the status 182 * for lock request, keep on trying 183 */ 184 for (;;) { 185 res = p9_client_lock_dotl(fid, &flock, &status); 186 if (res < 0) 187 goto out_unlock; 188 189 if (status != P9_LOCK_BLOCKED) 190 break; 191 if (status == P9_LOCK_BLOCKED && !IS_SETLKW(cmd)) 192 break; 193 if (schedule_timeout_interruptible(v9ses->session_lock_timeout) 194 != 0) 195 break; 196 /* 197 * p9_client_lock_dotl overwrites flock.client_id with the 198 * server message, free and reuse the client name 199 */ 200 if (flock.client_id != fid->clnt->name) { 201 kfree(flock.client_id); 202 flock.client_id = fid->clnt->name; 203 } 204 } 205 206 /* map 9p status to VFS status */ 207 switch (status) { 208 case P9_LOCK_SUCCESS: 209 res = 0; 210 break; 211 case P9_LOCK_BLOCKED: 212 res = -EAGAIN; 213 break; 214 default: 215 WARN_ONCE(1, "unknown lock status code: %d\n", status); 216 /* fall through */ 217 case P9_LOCK_ERROR: 218 case P9_LOCK_GRACE: 219 res = -ENOLCK; 220 break; 221 } 222 223 out_unlock: 224 /* 225 * incase server returned error for lock request, revert 226 * it locally 227 */ 228 if (res < 0 && fl->fl_type != F_UNLCK) { 229 fl_type = fl->fl_type; 230 fl->fl_type = F_UNLCK; 231 /* Even if this fails we want to return the remote error */ 232 locks_lock_file_wait(filp, fl); 233 fl->fl_type = fl_type; 234 } 235 if (flock.client_id != fid->clnt->name) 236 kfree(flock.client_id); 237 out: 238 return res; 239 } 240 241 static int v9fs_file_getlock(struct file *filp, struct file_lock *fl) 242 { 243 struct p9_getlock glock; 244 struct p9_fid *fid; 245 int res = 0; 246 247 fid = filp->private_data; 248 BUG_ON(fid == NULL); 249 250 posix_test_lock(filp, fl); 251 /* 252 * if we have a conflicting lock locally, no need to validate 253 * with server 254 */ 255 if (fl->fl_type != F_UNLCK) 256 return res; 257 258 /* convert posix lock to p9 tgetlock args */ 259 memset(&glock, 0, sizeof(glock)); 260 glock.type = P9_LOCK_TYPE_UNLCK; 261 glock.start = fl->fl_start; 262 if (fl->fl_end == OFFSET_MAX) 263 glock.length = 0; 264 else 265 glock.length = fl->fl_end - fl->fl_start + 1; 266 glock.proc_id = fl->fl_pid; 267 glock.client_id = fid->clnt->name; 268 269 res = p9_client_getlock_dotl(fid, &glock); 270 if (res < 0) 271 goto out; 272 /* map 9p lock type to os lock type */ 273 switch (glock.type) { 274 case P9_LOCK_TYPE_RDLCK: 275 fl->fl_type = F_RDLCK; 276 break; 277 case P9_LOCK_TYPE_WRLCK: 278 fl->fl_type = F_WRLCK; 279 break; 280 case P9_LOCK_TYPE_UNLCK: 281 fl->fl_type = F_UNLCK; 282 break; 283 } 284 if (glock.type != P9_LOCK_TYPE_UNLCK) { 285 fl->fl_start = glock.start; 286 if (glock.length == 0) 287 fl->fl_end = OFFSET_MAX; 288 else 289 fl->fl_end = glock.start + glock.length - 1; 290 fl->fl_pid = -glock.proc_id; 291 } 292 out: 293 if (glock.client_id != fid->clnt->name) 294 kfree(glock.client_id); 295 return res; 296 } 297 298 /** 299 * v9fs_file_lock_dotl - lock a file (or directory) 300 * @filp: file to be locked 301 * @cmd: lock command 302 * @fl: file lock structure 303 * 304 */ 305 306 static int v9fs_file_lock_dotl(struct file *filp, int cmd, struct file_lock *fl) 307 { 308 struct inode *inode = file_inode(filp); 309 int ret = -ENOLCK; 310 311 p9_debug(P9_DEBUG_VFS, "filp: %p cmd:%d lock: %p name: %pD\n", 312 filp, cmd, fl, filp); 313 314 /* No mandatory locks */ 315 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) 316 goto out_err; 317 318 if ((IS_SETLK(cmd) || IS_SETLKW(cmd)) && fl->fl_type != F_UNLCK) { 319 filemap_write_and_wait(inode->i_mapping); 320 invalidate_mapping_pages(&inode->i_data, 0, -1); 321 } 322 323 if (IS_SETLK(cmd) || IS_SETLKW(cmd)) 324 ret = v9fs_file_do_lock(filp, cmd, fl); 325 else if (IS_GETLK(cmd)) 326 ret = v9fs_file_getlock(filp, fl); 327 else 328 ret = -EINVAL; 329 out_err: 330 return ret; 331 } 332 333 /** 334 * v9fs_file_flock_dotl - lock a file 335 * @filp: file to be locked 336 * @cmd: lock command 337 * @fl: file lock structure 338 * 339 */ 340 341 static int v9fs_file_flock_dotl(struct file *filp, int cmd, 342 struct file_lock *fl) 343 { 344 struct inode *inode = file_inode(filp); 345 int ret = -ENOLCK; 346 347 p9_debug(P9_DEBUG_VFS, "filp: %p cmd:%d lock: %p name: %pD\n", 348 filp, cmd, fl, filp); 349 350 /* No mandatory locks */ 351 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) 352 goto out_err; 353 354 if (!(fl->fl_flags & FL_FLOCK)) 355 goto out_err; 356 357 if ((IS_SETLK(cmd) || IS_SETLKW(cmd)) && fl->fl_type != F_UNLCK) { 358 filemap_write_and_wait(inode->i_mapping); 359 invalidate_mapping_pages(&inode->i_data, 0, -1); 360 } 361 /* Convert flock to posix lock */ 362 fl->fl_flags |= FL_POSIX; 363 fl->fl_flags ^= FL_FLOCK; 364 365 if (IS_SETLK(cmd) | IS_SETLKW(cmd)) 366 ret = v9fs_file_do_lock(filp, cmd, fl); 367 else 368 ret = -EINVAL; 369 out_err: 370 return ret; 371 } 372 373 /** 374 * v9fs_file_read - read from a file 375 * @filp: file pointer to read 376 * @udata: user data buffer to read data into 377 * @count: size of buffer 378 * @offset: offset at which to read data 379 * 380 */ 381 382 static ssize_t 383 v9fs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 384 { 385 struct p9_fid *fid = iocb->ki_filp->private_data; 386 int ret, err = 0; 387 388 p9_debug(P9_DEBUG_VFS, "count %zu offset %lld\n", 389 iov_iter_count(to), iocb->ki_pos); 390 391 ret = p9_client_read(fid, iocb->ki_pos, to, &err); 392 if (!ret) 393 return err; 394 395 iocb->ki_pos += ret; 396 return ret; 397 } 398 399 /** 400 * v9fs_file_write - write to a file 401 * @filp: file pointer to write 402 * @data: data buffer to write data from 403 * @count: size of buffer 404 * @offset: offset at which to write data 405 * 406 */ 407 static ssize_t 408 v9fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 409 { 410 struct file *file = iocb->ki_filp; 411 ssize_t retval; 412 loff_t origin; 413 int err = 0; 414 415 retval = generic_write_checks(iocb, from); 416 if (retval <= 0) 417 return retval; 418 419 origin = iocb->ki_pos; 420 retval = p9_client_write(file->private_data, iocb->ki_pos, from, &err); 421 if (retval > 0) { 422 struct inode *inode = file_inode(file); 423 loff_t i_size; 424 unsigned long pg_start, pg_end; 425 pg_start = origin >> PAGE_SHIFT; 426 pg_end = (origin + retval - 1) >> PAGE_SHIFT; 427 if (inode->i_mapping && inode->i_mapping->nrpages) 428 invalidate_inode_pages2_range(inode->i_mapping, 429 pg_start, pg_end); 430 iocb->ki_pos += retval; 431 i_size = i_size_read(inode); 432 if (iocb->ki_pos > i_size) { 433 inode_add_bytes(inode, iocb->ki_pos - i_size); 434 /* 435 * Need to serialize against i_size_write() in 436 * v9fs_stat2inode() 437 */ 438 v9fs_i_size_write(inode, iocb->ki_pos); 439 } 440 return retval; 441 } 442 return err; 443 } 444 445 static int v9fs_file_fsync(struct file *filp, loff_t start, loff_t end, 446 int datasync) 447 { 448 struct p9_fid *fid; 449 struct inode *inode = filp->f_mapping->host; 450 struct p9_wstat wstat; 451 int retval; 452 453 retval = file_write_and_wait_range(filp, start, end); 454 if (retval) 455 return retval; 456 457 inode_lock(inode); 458 p9_debug(P9_DEBUG_VFS, "filp %p datasync %x\n", filp, datasync); 459 460 fid = filp->private_data; 461 v9fs_blank_wstat(&wstat); 462 463 retval = p9_client_wstat(fid, &wstat); 464 inode_unlock(inode); 465 466 return retval; 467 } 468 469 int v9fs_file_fsync_dotl(struct file *filp, loff_t start, loff_t end, 470 int datasync) 471 { 472 struct p9_fid *fid; 473 struct inode *inode = filp->f_mapping->host; 474 int retval; 475 476 retval = file_write_and_wait_range(filp, start, end); 477 if (retval) 478 return retval; 479 480 inode_lock(inode); 481 p9_debug(P9_DEBUG_VFS, "filp %p datasync %x\n", filp, datasync); 482 483 fid = filp->private_data; 484 485 retval = p9_client_fsync(fid, datasync); 486 inode_unlock(inode); 487 488 return retval; 489 } 490 491 static int 492 v9fs_file_mmap(struct file *filp, struct vm_area_struct *vma) 493 { 494 int retval; 495 496 497 retval = generic_file_mmap(filp, vma); 498 if (!retval) 499 vma->vm_ops = &v9fs_file_vm_ops; 500 501 return retval; 502 } 503 504 static int 505 v9fs_mmap_file_mmap(struct file *filp, struct vm_area_struct *vma) 506 { 507 int retval; 508 struct inode *inode; 509 struct v9fs_inode *v9inode; 510 struct p9_fid *fid; 511 512 inode = file_inode(filp); 513 v9inode = V9FS_I(inode); 514 mutex_lock(&v9inode->v_mutex); 515 if (!v9inode->writeback_fid && 516 (vma->vm_flags & VM_SHARED) && 517 (vma->vm_flags & VM_WRITE)) { 518 /* 519 * clone a fid and add it to writeback_fid 520 * we do it during mmap instead of 521 * page dirty time via write_begin/page_mkwrite 522 * because we want write after unlink usecase 523 * to work. 524 */ 525 fid = v9fs_writeback_fid(file_dentry(filp)); 526 if (IS_ERR(fid)) { 527 retval = PTR_ERR(fid); 528 mutex_unlock(&v9inode->v_mutex); 529 return retval; 530 } 531 v9inode->writeback_fid = (void *) fid; 532 } 533 mutex_unlock(&v9inode->v_mutex); 534 535 retval = generic_file_mmap(filp, vma); 536 if (!retval) 537 vma->vm_ops = &v9fs_mmap_file_vm_ops; 538 539 return retval; 540 } 541 542 static vm_fault_t 543 v9fs_vm_page_mkwrite(struct vm_fault *vmf) 544 { 545 struct v9fs_inode *v9inode; 546 struct page *page = vmf->page; 547 struct file *filp = vmf->vma->vm_file; 548 struct inode *inode = file_inode(filp); 549 550 551 p9_debug(P9_DEBUG_VFS, "page %p fid %lx\n", 552 page, (unsigned long)filp->private_data); 553 554 /* Update file times before taking page lock */ 555 file_update_time(filp); 556 557 v9inode = V9FS_I(inode); 558 /* make sure the cache has finished storing the page */ 559 v9fs_fscache_wait_on_page_write(inode, page); 560 BUG_ON(!v9inode->writeback_fid); 561 lock_page(page); 562 if (page->mapping != inode->i_mapping) 563 goto out_unlock; 564 wait_for_stable_page(page); 565 566 return VM_FAULT_LOCKED; 567 out_unlock: 568 unlock_page(page); 569 return VM_FAULT_NOPAGE; 570 } 571 572 /** 573 * v9fs_mmap_file_read - read from a file 574 * @filp: file pointer to read 575 * @data: user data buffer to read data into 576 * @count: size of buffer 577 * @offset: offset at which to read data 578 * 579 */ 580 static ssize_t 581 v9fs_mmap_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 582 { 583 /* TODO: Check if there are dirty pages */ 584 return v9fs_file_read_iter(iocb, to); 585 } 586 587 /** 588 * v9fs_mmap_file_write - write to a file 589 * @filp: file pointer to write 590 * @data: data buffer to write data from 591 * @count: size of buffer 592 * @offset: offset at which to write data 593 * 594 */ 595 static ssize_t 596 v9fs_mmap_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 597 { 598 /* 599 * TODO: invalidate mmaps on filp's inode between 600 * offset and offset+count 601 */ 602 return v9fs_file_write_iter(iocb, from); 603 } 604 605 static void v9fs_mmap_vm_close(struct vm_area_struct *vma) 606 { 607 struct inode *inode; 608 609 struct writeback_control wbc = { 610 .nr_to_write = LONG_MAX, 611 .sync_mode = WB_SYNC_ALL, 612 .range_start = vma->vm_pgoff * PAGE_SIZE, 613 /* absolute end, byte at end included */ 614 .range_end = vma->vm_pgoff * PAGE_SIZE + 615 (vma->vm_end - vma->vm_start - 1), 616 }; 617 618 if (!(vma->vm_flags & VM_SHARED)) 619 return; 620 621 p9_debug(P9_DEBUG_VFS, "9p VMA close, %p, flushing", vma); 622 623 inode = file_inode(vma->vm_file); 624 625 if (!mapping_cap_writeback_dirty(inode->i_mapping)) 626 wbc.nr_to_write = 0; 627 628 might_sleep(); 629 sync_inode(inode, &wbc); 630 } 631 632 633 static const struct vm_operations_struct v9fs_file_vm_ops = { 634 .fault = filemap_fault, 635 .map_pages = filemap_map_pages, 636 .page_mkwrite = v9fs_vm_page_mkwrite, 637 }; 638 639 static const struct vm_operations_struct v9fs_mmap_file_vm_ops = { 640 .close = v9fs_mmap_vm_close, 641 .fault = filemap_fault, 642 .map_pages = filemap_map_pages, 643 .page_mkwrite = v9fs_vm_page_mkwrite, 644 }; 645 646 647 const struct file_operations v9fs_cached_file_operations = { 648 .llseek = generic_file_llseek, 649 .read_iter = generic_file_read_iter, 650 .write_iter = generic_file_write_iter, 651 .open = v9fs_file_open, 652 .release = v9fs_dir_release, 653 .lock = v9fs_file_lock, 654 .mmap = v9fs_file_mmap, 655 .fsync = v9fs_file_fsync, 656 }; 657 658 const struct file_operations v9fs_cached_file_operations_dotl = { 659 .llseek = generic_file_llseek, 660 .read_iter = generic_file_read_iter, 661 .write_iter = generic_file_write_iter, 662 .open = v9fs_file_open, 663 .release = v9fs_dir_release, 664 .lock = v9fs_file_lock_dotl, 665 .flock = v9fs_file_flock_dotl, 666 .mmap = v9fs_file_mmap, 667 .fsync = v9fs_file_fsync_dotl, 668 }; 669 670 const struct file_operations v9fs_file_operations = { 671 .llseek = generic_file_llseek, 672 .read_iter = v9fs_file_read_iter, 673 .write_iter = v9fs_file_write_iter, 674 .open = v9fs_file_open, 675 .release = v9fs_dir_release, 676 .lock = v9fs_file_lock, 677 .mmap = generic_file_readonly_mmap, 678 .fsync = v9fs_file_fsync, 679 }; 680 681 const struct file_operations v9fs_file_operations_dotl = { 682 .llseek = generic_file_llseek, 683 .read_iter = v9fs_file_read_iter, 684 .write_iter = v9fs_file_write_iter, 685 .open = v9fs_file_open, 686 .release = v9fs_dir_release, 687 .lock = v9fs_file_lock_dotl, 688 .flock = v9fs_file_flock_dotl, 689 .mmap = generic_file_readonly_mmap, 690 .fsync = v9fs_file_fsync_dotl, 691 }; 692 693 const struct file_operations v9fs_mmap_file_operations = { 694 .llseek = generic_file_llseek, 695 .read_iter = v9fs_mmap_file_read_iter, 696 .write_iter = v9fs_mmap_file_write_iter, 697 .open = v9fs_file_open, 698 .release = v9fs_dir_release, 699 .lock = v9fs_file_lock, 700 .mmap = v9fs_mmap_file_mmap, 701 .fsync = v9fs_file_fsync, 702 }; 703 704 const struct file_operations v9fs_mmap_file_operations_dotl = { 705 .llseek = generic_file_llseek, 706 .read_iter = v9fs_mmap_file_read_iter, 707 .write_iter = v9fs_mmap_file_write_iter, 708 .open = v9fs_file_open, 709 .release = v9fs_dir_release, 710 .lock = v9fs_file_lock_dotl, 711 .flock = v9fs_file_flock_dotl, 712 .mmap = v9fs_mmap_file_mmap, 713 .fsync = v9fs_file_fsync_dotl, 714 }; 715