1 #include <linux/ceph/ceph_debug.h> 2 3 #include <linux/module.h> 4 #include <linux/sched.h> 5 #include <linux/slab.h> 6 #include <linux/file.h> 7 #include <linux/mount.h> 8 #include <linux/namei.h> 9 #include <linux/writeback.h> 10 #include <linux/falloc.h> 11 12 #include "super.h" 13 #include "mds_client.h" 14 #include "cache.h" 15 16 /* 17 * Ceph file operations 18 * 19 * Implement basic open/close functionality, and implement 20 * read/write. 21 * 22 * We implement three modes of file I/O: 23 * - buffered uses the generic_file_aio_{read,write} helpers 24 * 25 * - synchronous is used when there is multi-client read/write 26 * sharing, avoids the page cache, and synchronously waits for an 27 * ack from the OSD. 28 * 29 * - direct io takes the variant of the sync path that references 30 * user pages directly. 31 * 32 * fsync() flushes and waits on dirty pages, but just queues metadata 33 * for writeback: since the MDS can recover size and mtime there is no 34 * need to wait for MDS acknowledgement. 35 */ 36 37 38 /* 39 * Prepare an open request. Preallocate ceph_cap to avoid an 40 * inopportune ENOMEM later. 41 */ 42 static struct ceph_mds_request * 43 prepare_open_request(struct super_block *sb, int flags, int create_mode) 44 { 45 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 46 struct ceph_mds_client *mdsc = fsc->mdsc; 47 struct ceph_mds_request *req; 48 int want_auth = USE_ANY_MDS; 49 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; 50 51 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) 52 want_auth = USE_AUTH_MDS; 53 54 req = ceph_mdsc_create_request(mdsc, op, want_auth); 55 if (IS_ERR(req)) 56 goto out; 57 req->r_fmode = ceph_flags_to_mode(flags); 58 req->r_args.open.flags = cpu_to_le32(flags); 59 req->r_args.open.mode = cpu_to_le32(create_mode); 60 out: 61 return req; 62 } 63 64 /* 65 * initialize private struct file data. 66 * if we fail, clean up by dropping fmode reference on the ceph_inode 67 */ 68 static int ceph_init_file(struct inode *inode, struct file *file, int fmode) 69 { 70 struct ceph_file_info *cf; 71 int ret = 0; 72 struct ceph_inode_info *ci = ceph_inode(inode); 73 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 74 struct ceph_mds_client *mdsc = fsc->mdsc; 75 76 switch (inode->i_mode & S_IFMT) { 77 case S_IFREG: 78 /* First file open request creates the cookie, we want to keep 79 * this cookie around for the filetime of the inode as not to 80 * have to worry about fscache register / revoke / operation 81 * races. 82 * 83 * Also, if we know the operation is going to invalidate data 84 * (non readonly) just nuke the cache right away. 85 */ 86 ceph_fscache_register_inode_cookie(mdsc->fsc, ci); 87 if ((fmode & CEPH_FILE_MODE_WR)) 88 ceph_fscache_invalidate(inode); 89 case S_IFDIR: 90 dout("init_file %p %p 0%o (regular)\n", inode, file, 91 inode->i_mode); 92 cf = kmem_cache_alloc(ceph_file_cachep, GFP_KERNEL | __GFP_ZERO); 93 if (cf == NULL) { 94 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ 95 return -ENOMEM; 96 } 97 cf->fmode = fmode; 98 cf->next_offset = 2; 99 cf->readdir_cache_idx = -1; 100 file->private_data = cf; 101 BUG_ON(inode->i_fop->release != ceph_release); 102 break; 103 104 case S_IFLNK: 105 dout("init_file %p %p 0%o (symlink)\n", inode, file, 106 inode->i_mode); 107 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ 108 break; 109 110 default: 111 dout("init_file %p %p 0%o (special)\n", inode, file, 112 inode->i_mode); 113 /* 114 * we need to drop the open ref now, since we don't 115 * have .release set to ceph_release. 116 */ 117 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ 118 BUG_ON(inode->i_fop->release == ceph_release); 119 120 /* call the proper open fop */ 121 ret = inode->i_fop->open(inode, file); 122 } 123 return ret; 124 } 125 126 /* 127 * If we already have the requisite capabilities, we can satisfy 128 * the open request locally (no need to request new caps from the 129 * MDS). We do, however, need to inform the MDS (asynchronously) 130 * if our wanted caps set expands. 131 */ 132 int ceph_open(struct inode *inode, struct file *file) 133 { 134 struct ceph_inode_info *ci = ceph_inode(inode); 135 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 136 struct ceph_mds_client *mdsc = fsc->mdsc; 137 struct ceph_mds_request *req; 138 struct ceph_file_info *cf = file->private_data; 139 struct inode *parent_inode = NULL; 140 int err; 141 int flags, fmode, wanted; 142 143 if (cf) { 144 dout("open file %p is already opened\n", file); 145 return 0; 146 } 147 148 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ 149 flags = file->f_flags & ~(O_CREAT|O_EXCL); 150 if (S_ISDIR(inode->i_mode)) 151 flags = O_DIRECTORY; /* mds likes to know */ 152 153 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, 154 ceph_vinop(inode), file, flags, file->f_flags); 155 fmode = ceph_flags_to_mode(flags); 156 wanted = ceph_caps_for_mode(fmode); 157 158 /* snapped files are read-only */ 159 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) 160 return -EROFS; 161 162 /* trivially open snapdir */ 163 if (ceph_snap(inode) == CEPH_SNAPDIR) { 164 spin_lock(&ci->i_ceph_lock); 165 __ceph_get_fmode(ci, fmode); 166 spin_unlock(&ci->i_ceph_lock); 167 return ceph_init_file(inode, file, fmode); 168 } 169 170 /* 171 * No need to block if we have caps on the auth MDS (for 172 * write) or any MDS (for read). Update wanted set 173 * asynchronously. 174 */ 175 spin_lock(&ci->i_ceph_lock); 176 if (__ceph_is_any_real_caps(ci) && 177 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { 178 int mds_wanted = __ceph_caps_mds_wanted(ci); 179 int issued = __ceph_caps_issued(ci, NULL); 180 181 dout("open %p fmode %d want %s issued %s using existing\n", 182 inode, fmode, ceph_cap_string(wanted), 183 ceph_cap_string(issued)); 184 __ceph_get_fmode(ci, fmode); 185 spin_unlock(&ci->i_ceph_lock); 186 187 /* adjust wanted? */ 188 if ((issued & wanted) != wanted && 189 (mds_wanted & wanted) != wanted && 190 ceph_snap(inode) != CEPH_SNAPDIR) 191 ceph_check_caps(ci, 0, NULL); 192 193 return ceph_init_file(inode, file, fmode); 194 } else if (ceph_snap(inode) != CEPH_NOSNAP && 195 (ci->i_snap_caps & wanted) == wanted) { 196 __ceph_get_fmode(ci, fmode); 197 spin_unlock(&ci->i_ceph_lock); 198 return ceph_init_file(inode, file, fmode); 199 } 200 201 spin_unlock(&ci->i_ceph_lock); 202 203 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); 204 req = prepare_open_request(inode->i_sb, flags, 0); 205 if (IS_ERR(req)) { 206 err = PTR_ERR(req); 207 goto out; 208 } 209 req->r_inode = inode; 210 ihold(inode); 211 212 req->r_num_caps = 1; 213 if (flags & O_CREAT) 214 parent_inode = ceph_get_dentry_parent_inode(file->f_path.dentry); 215 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 216 iput(parent_inode); 217 if (!err) 218 err = ceph_init_file(inode, file, req->r_fmode); 219 ceph_mdsc_put_request(req); 220 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); 221 out: 222 return err; 223 } 224 225 226 /* 227 * Do a lookup + open with a single request. If we get a non-existent 228 * file or symlink, return 1 so the VFS can retry. 229 */ 230 int ceph_atomic_open(struct inode *dir, struct dentry *dentry, 231 struct file *file, unsigned flags, umode_t mode, 232 int *opened) 233 { 234 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); 235 struct ceph_mds_client *mdsc = fsc->mdsc; 236 struct ceph_mds_request *req; 237 struct dentry *dn; 238 struct ceph_acls_info acls = {}; 239 int err; 240 241 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", 242 dir, dentry, dentry, 243 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); 244 245 if (dentry->d_name.len > NAME_MAX) 246 return -ENAMETOOLONG; 247 248 err = ceph_init_dentry(dentry); 249 if (err < 0) 250 return err; 251 252 if (flags & O_CREAT) { 253 err = ceph_pre_init_acls(dir, &mode, &acls); 254 if (err < 0) 255 return err; 256 } 257 258 /* do the open */ 259 req = prepare_open_request(dir->i_sb, flags, mode); 260 if (IS_ERR(req)) { 261 err = PTR_ERR(req); 262 goto out_acl; 263 } 264 req->r_dentry = dget(dentry); 265 req->r_num_caps = 2; 266 if (flags & O_CREAT) { 267 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 268 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 269 if (acls.pagelist) { 270 req->r_pagelist = acls.pagelist; 271 acls.pagelist = NULL; 272 } 273 } 274 req->r_locked_dir = dir; /* caller holds dir->i_mutex */ 275 err = ceph_mdsc_do_request(mdsc, 276 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, 277 req); 278 err = ceph_handle_snapdir(req, dentry, err); 279 if (err) 280 goto out_req; 281 282 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry) 283 err = ceph_handle_notrace_create(dir, dentry); 284 285 if (d_unhashed(dentry)) { 286 dn = ceph_finish_lookup(req, dentry, err); 287 if (IS_ERR(dn)) 288 err = PTR_ERR(dn); 289 } else { 290 /* we were given a hashed negative dentry */ 291 dn = NULL; 292 } 293 if (err) 294 goto out_req; 295 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { 296 /* make vfs retry on splice, ENOENT, or symlink */ 297 dout("atomic_open finish_no_open on dn %p\n", dn); 298 err = finish_no_open(file, dn); 299 } else { 300 dout("atomic_open finish_open on dn %p\n", dn); 301 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { 302 ceph_init_inode_acls(d_inode(dentry), &acls); 303 *opened |= FILE_CREATED; 304 } 305 err = finish_open(file, dentry, ceph_open, opened); 306 } 307 out_req: 308 if (!req->r_err && req->r_target_inode) 309 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); 310 ceph_mdsc_put_request(req); 311 out_acl: 312 ceph_release_acls_info(&acls); 313 dout("atomic_open result=%d\n", err); 314 return err; 315 } 316 317 int ceph_release(struct inode *inode, struct file *file) 318 { 319 struct ceph_inode_info *ci = ceph_inode(inode); 320 struct ceph_file_info *cf = file->private_data; 321 322 dout("release inode %p file %p\n", inode, file); 323 ceph_put_fmode(ci, cf->fmode); 324 if (cf->last_readdir) 325 ceph_mdsc_put_request(cf->last_readdir); 326 kfree(cf->last_name); 327 kfree(cf->dir_info); 328 kmem_cache_free(ceph_file_cachep, cf); 329 330 /* wake up anyone waiting for caps on this inode */ 331 wake_up_all(&ci->i_cap_wq); 332 return 0; 333 } 334 335 enum { 336 CHECK_EOF = 1, 337 READ_INLINE = 2, 338 }; 339 340 /* 341 * Read a range of bytes striped over one or more objects. Iterate over 342 * objects we stripe over. (That's not atomic, but good enough for now.) 343 * 344 * If we get a short result from the OSD, check against i_size; we need to 345 * only return a short read to the caller if we hit EOF. 346 */ 347 static int striped_read(struct inode *inode, 348 u64 off, u64 len, 349 struct page **pages, int num_pages, 350 int *checkeof, bool o_direct, 351 unsigned long buf_align) 352 { 353 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 354 struct ceph_inode_info *ci = ceph_inode(inode); 355 u64 pos, this_len, left; 356 int io_align, page_align; 357 int pages_left; 358 int read; 359 struct page **page_pos; 360 int ret; 361 bool hit_stripe, was_short; 362 363 /* 364 * we may need to do multiple reads. not atomic, unfortunately. 365 */ 366 pos = off; 367 left = len; 368 page_pos = pages; 369 pages_left = num_pages; 370 read = 0; 371 io_align = off & ~PAGE_MASK; 372 373 more: 374 if (o_direct) 375 page_align = (pos - io_align + buf_align) & ~PAGE_MASK; 376 else 377 page_align = pos & ~PAGE_MASK; 378 this_len = left; 379 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), 380 &ci->i_layout, pos, &this_len, 381 ci->i_truncate_seq, 382 ci->i_truncate_size, 383 page_pos, pages_left, page_align); 384 if (ret == -ENOENT) 385 ret = 0; 386 hit_stripe = this_len < left; 387 was_short = ret >= 0 && ret < this_len; 388 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, 389 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); 390 391 if (ret >= 0) { 392 int didpages; 393 if (was_short && (pos + ret < inode->i_size)) { 394 int zlen = min(this_len - ret, 395 inode->i_size - pos - ret); 396 int zoff = (o_direct ? buf_align : io_align) + 397 read + ret; 398 dout(" zero gap %llu to %llu\n", 399 pos + ret, pos + ret + zlen); 400 ceph_zero_page_vector_range(zoff, zlen, pages); 401 ret += zlen; 402 } 403 404 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; 405 pos += ret; 406 read = pos - off; 407 left -= ret; 408 page_pos += didpages; 409 pages_left -= didpages; 410 411 /* hit stripe and need continue*/ 412 if (left && hit_stripe && pos < inode->i_size) 413 goto more; 414 } 415 416 if (read > 0) { 417 ret = read; 418 /* did we bounce off eof? */ 419 if (pos + left > inode->i_size) 420 *checkeof = CHECK_EOF; 421 } 422 423 dout("striped_read returns %d\n", ret); 424 return ret; 425 } 426 427 /* 428 * Completely synchronous read and write methods. Direct from __user 429 * buffer to osd, or directly to user pages (if O_DIRECT). 430 * 431 * If the read spans object boundary, just do multiple reads. 432 */ 433 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, 434 int *checkeof) 435 { 436 struct file *file = iocb->ki_filp; 437 struct inode *inode = file_inode(file); 438 struct page **pages; 439 u64 off = iocb->ki_pos; 440 int num_pages, ret; 441 size_t len = iov_iter_count(i); 442 443 dout("sync_read on file %p %llu~%u %s\n", file, off, 444 (unsigned)len, 445 (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); 446 447 if (!len) 448 return 0; 449 /* 450 * flush any page cache pages in this range. this 451 * will make concurrent normal and sync io slow, 452 * but it will at least behave sensibly when they are 453 * in sequence. 454 */ 455 ret = filemap_write_and_wait_range(inode->i_mapping, off, 456 off + len); 457 if (ret < 0) 458 return ret; 459 460 if (iocb->ki_flags & IOCB_DIRECT) { 461 while (iov_iter_count(i)) { 462 size_t start; 463 ssize_t n; 464 465 n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start); 466 if (n < 0) 467 return n; 468 469 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE; 470 471 ret = striped_read(inode, off, n, 472 pages, num_pages, checkeof, 473 1, start); 474 475 ceph_put_page_vector(pages, num_pages, true); 476 477 if (ret <= 0) 478 break; 479 off += ret; 480 iov_iter_advance(i, ret); 481 if (ret < n) 482 break; 483 } 484 } else { 485 num_pages = calc_pages_for(off, len); 486 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); 487 if (IS_ERR(pages)) 488 return PTR_ERR(pages); 489 ret = striped_read(inode, off, len, pages, 490 num_pages, checkeof, 0, 0); 491 if (ret > 0) { 492 int l, k = 0; 493 size_t left = ret; 494 495 while (left) { 496 size_t page_off = off & ~PAGE_MASK; 497 size_t copy = min_t(size_t, 498 PAGE_SIZE - page_off, left); 499 l = copy_page_to_iter(pages[k++], page_off, 500 copy, i); 501 off += l; 502 left -= l; 503 if (l < copy) 504 break; 505 } 506 } 507 ceph_release_page_vector(pages, num_pages); 508 } 509 510 if (off > iocb->ki_pos) { 511 ret = off - iocb->ki_pos; 512 iocb->ki_pos = off; 513 } 514 515 dout("sync_read result %d\n", ret); 516 return ret; 517 } 518 519 /* 520 * Write commit request unsafe callback, called to tell us when a 521 * request is unsafe (that is, in flight--has been handed to the 522 * messenger to send to its target osd). It is called again when 523 * we've received a response message indicating the request is 524 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request 525 * is completed early (and unsuccessfully) due to a timeout or 526 * interrupt. 527 * 528 * This is used if we requested both an ACK and ONDISK commit reply 529 * from the OSD. 530 */ 531 static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) 532 { 533 struct ceph_inode_info *ci = ceph_inode(req->r_inode); 534 535 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, 536 unsafe ? "un" : ""); 537 if (unsafe) { 538 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); 539 spin_lock(&ci->i_unsafe_lock); 540 list_add_tail(&req->r_unsafe_item, 541 &ci->i_unsafe_writes); 542 spin_unlock(&ci->i_unsafe_lock); 543 } else { 544 spin_lock(&ci->i_unsafe_lock); 545 list_del_init(&req->r_unsafe_item); 546 spin_unlock(&ci->i_unsafe_lock); 547 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); 548 } 549 } 550 551 552 /* 553 * Synchronous write, straight from __user pointer or user pages. 554 * 555 * If write spans object boundary, just do multiple writes. (For a 556 * correct atomic write, we should e.g. take write locks on all 557 * objects, rollback on failure, etc.) 558 */ 559 static ssize_t 560 ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, 561 struct ceph_snap_context *snapc) 562 { 563 struct file *file = iocb->ki_filp; 564 struct inode *inode = file_inode(file); 565 struct ceph_inode_info *ci = ceph_inode(inode); 566 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 567 struct ceph_vino vino; 568 struct ceph_osd_request *req; 569 struct page **pages; 570 int num_pages; 571 int written = 0; 572 int flags; 573 int check_caps = 0; 574 int ret; 575 struct timespec mtime = CURRENT_TIME; 576 size_t count = iov_iter_count(from); 577 578 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) 579 return -EROFS; 580 581 dout("sync_direct_write on file %p %lld~%u\n", file, pos, 582 (unsigned)count); 583 584 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); 585 if (ret < 0) 586 return ret; 587 588 ret = invalidate_inode_pages2_range(inode->i_mapping, 589 pos >> PAGE_CACHE_SHIFT, 590 (pos + count) >> PAGE_CACHE_SHIFT); 591 if (ret < 0) 592 dout("invalidate_inode_pages2_range returned %d\n", ret); 593 594 flags = CEPH_OSD_FLAG_ORDERSNAP | 595 CEPH_OSD_FLAG_ONDISK | 596 CEPH_OSD_FLAG_WRITE; 597 598 while (iov_iter_count(from) > 0) { 599 u64 len = iov_iter_single_seg_count(from); 600 size_t start; 601 ssize_t n; 602 603 vino = ceph_vino(inode); 604 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 605 vino, pos, &len, 0, 606 2,/*include a 'startsync' command*/ 607 CEPH_OSD_OP_WRITE, flags, snapc, 608 ci->i_truncate_seq, 609 ci->i_truncate_size, 610 false); 611 if (IS_ERR(req)) { 612 ret = PTR_ERR(req); 613 break; 614 } 615 616 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); 617 618 n = iov_iter_get_pages_alloc(from, &pages, len, &start); 619 if (unlikely(n < 0)) { 620 ret = n; 621 ceph_osdc_put_request(req); 622 break; 623 } 624 625 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE; 626 /* 627 * throw out any page cache pages in this range. this 628 * may block. 629 */ 630 truncate_inode_pages_range(inode->i_mapping, pos, 631 (pos+n) | (PAGE_CACHE_SIZE-1)); 632 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start, 633 false, false); 634 635 /* BUG_ON(vino.snap != CEPH_NOSNAP); */ 636 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); 637 638 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); 639 if (!ret) 640 ret = ceph_osdc_wait_request(&fsc->client->osdc, req); 641 642 ceph_put_page_vector(pages, num_pages, false); 643 644 ceph_osdc_put_request(req); 645 if (ret) 646 break; 647 pos += n; 648 written += n; 649 iov_iter_advance(from, n); 650 651 if (pos > i_size_read(inode)) { 652 check_caps = ceph_inode_set_size(inode, pos); 653 if (check_caps) 654 ceph_check_caps(ceph_inode(inode), 655 CHECK_CAPS_AUTHONLY, 656 NULL); 657 } 658 } 659 660 if (ret != -EOLDSNAPC && written > 0) { 661 iocb->ki_pos = pos; 662 ret = written; 663 } 664 return ret; 665 } 666 667 668 /* 669 * Synchronous write, straight from __user pointer or user pages. 670 * 671 * If write spans object boundary, just do multiple writes. (For a 672 * correct atomic write, we should e.g. take write locks on all 673 * objects, rollback on failure, etc.) 674 */ 675 static ssize_t 676 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, 677 struct ceph_snap_context *snapc) 678 { 679 struct file *file = iocb->ki_filp; 680 struct inode *inode = file_inode(file); 681 struct ceph_inode_info *ci = ceph_inode(inode); 682 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 683 struct ceph_vino vino; 684 struct ceph_osd_request *req; 685 struct page **pages; 686 u64 len; 687 int num_pages; 688 int written = 0; 689 int flags; 690 int check_caps = 0; 691 int ret; 692 struct timespec mtime = CURRENT_TIME; 693 size_t count = iov_iter_count(from); 694 695 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) 696 return -EROFS; 697 698 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); 699 700 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); 701 if (ret < 0) 702 return ret; 703 704 ret = invalidate_inode_pages2_range(inode->i_mapping, 705 pos >> PAGE_CACHE_SHIFT, 706 (pos + count) >> PAGE_CACHE_SHIFT); 707 if (ret < 0) 708 dout("invalidate_inode_pages2_range returned %d\n", ret); 709 710 flags = CEPH_OSD_FLAG_ORDERSNAP | 711 CEPH_OSD_FLAG_ONDISK | 712 CEPH_OSD_FLAG_WRITE | 713 CEPH_OSD_FLAG_ACK; 714 715 while ((len = iov_iter_count(from)) > 0) { 716 size_t left; 717 int n; 718 719 vino = ceph_vino(inode); 720 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 721 vino, pos, &len, 0, 1, 722 CEPH_OSD_OP_WRITE, flags, snapc, 723 ci->i_truncate_seq, 724 ci->i_truncate_size, 725 false); 726 if (IS_ERR(req)) { 727 ret = PTR_ERR(req); 728 break; 729 } 730 731 /* 732 * write from beginning of first page, 733 * regardless of io alignment 734 */ 735 num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 736 737 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); 738 if (IS_ERR(pages)) { 739 ret = PTR_ERR(pages); 740 goto out; 741 } 742 743 left = len; 744 for (n = 0; n < num_pages; n++) { 745 size_t plen = min_t(size_t, left, PAGE_SIZE); 746 ret = copy_page_from_iter(pages[n], 0, plen, from); 747 if (ret != plen) { 748 ret = -EFAULT; 749 break; 750 } 751 left -= ret; 752 } 753 754 if (ret < 0) { 755 ceph_release_page_vector(pages, num_pages); 756 goto out; 757 } 758 759 /* get a second commit callback */ 760 req->r_unsafe_callback = ceph_sync_write_unsafe; 761 req->r_inode = inode; 762 763 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, 764 false, true); 765 766 /* BUG_ON(vino.snap != CEPH_NOSNAP); */ 767 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); 768 769 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); 770 if (!ret) 771 ret = ceph_osdc_wait_request(&fsc->client->osdc, req); 772 773 out: 774 ceph_osdc_put_request(req); 775 if (ret == 0) { 776 pos += len; 777 written += len; 778 779 if (pos > i_size_read(inode)) { 780 check_caps = ceph_inode_set_size(inode, pos); 781 if (check_caps) 782 ceph_check_caps(ceph_inode(inode), 783 CHECK_CAPS_AUTHONLY, 784 NULL); 785 } 786 } else 787 break; 788 } 789 790 if (ret != -EOLDSNAPC && written > 0) { 791 ret = written; 792 iocb->ki_pos = pos; 793 } 794 return ret; 795 } 796 797 /* 798 * Wrap generic_file_aio_read with checks for cap bits on the inode. 799 * Atomically grab references, so that those bits are not released 800 * back to the MDS mid-read. 801 * 802 * Hmm, the sync read case isn't actually async... should it be? 803 */ 804 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) 805 { 806 struct file *filp = iocb->ki_filp; 807 struct ceph_file_info *fi = filp->private_data; 808 size_t len = iov_iter_count(to); 809 struct inode *inode = file_inode(filp); 810 struct ceph_inode_info *ci = ceph_inode(inode); 811 struct page *pinned_page = NULL; 812 ssize_t ret; 813 int want, got = 0; 814 int retry_op = 0, read = 0; 815 816 again: 817 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", 818 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); 819 820 if (fi->fmode & CEPH_FILE_MODE_LAZY) 821 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 822 else 823 want = CEPH_CAP_FILE_CACHE; 824 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); 825 if (ret < 0) 826 return ret; 827 828 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || 829 (iocb->ki_flags & IOCB_DIRECT) || 830 (fi->flags & CEPH_F_SYNC)) { 831 832 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", 833 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, 834 ceph_cap_string(got)); 835 836 if (ci->i_inline_version == CEPH_INLINE_NONE) { 837 /* hmm, this isn't really async... */ 838 ret = ceph_sync_read(iocb, to, &retry_op); 839 } else { 840 retry_op = READ_INLINE; 841 } 842 } else { 843 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", 844 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, 845 ceph_cap_string(got)); 846 847 ret = generic_file_read_iter(iocb, to); 848 } 849 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", 850 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); 851 if (pinned_page) { 852 page_cache_release(pinned_page); 853 pinned_page = NULL; 854 } 855 ceph_put_cap_refs(ci, got); 856 if (retry_op && ret >= 0) { 857 int statret; 858 struct page *page = NULL; 859 loff_t i_size; 860 if (retry_op == READ_INLINE) { 861 page = __page_cache_alloc(GFP_KERNEL); 862 if (!page) 863 return -ENOMEM; 864 } 865 866 statret = __ceph_do_getattr(inode, page, 867 CEPH_STAT_CAP_INLINE_DATA, !!page); 868 if (statret < 0) { 869 __free_page(page); 870 if (statret == -ENODATA) { 871 BUG_ON(retry_op != READ_INLINE); 872 goto again; 873 } 874 return statret; 875 } 876 877 i_size = i_size_read(inode); 878 if (retry_op == READ_INLINE) { 879 BUG_ON(ret > 0 || read > 0); 880 if (iocb->ki_pos < i_size && 881 iocb->ki_pos < PAGE_CACHE_SIZE) { 882 loff_t end = min_t(loff_t, i_size, 883 iocb->ki_pos + len); 884 end = min_t(loff_t, end, PAGE_CACHE_SIZE); 885 if (statret < end) 886 zero_user_segment(page, statret, end); 887 ret = copy_page_to_iter(page, 888 iocb->ki_pos & ~PAGE_MASK, 889 end - iocb->ki_pos, to); 890 iocb->ki_pos += ret; 891 read += ret; 892 } 893 if (iocb->ki_pos < i_size && read < len) { 894 size_t zlen = min_t(size_t, len - read, 895 i_size - iocb->ki_pos); 896 ret = iov_iter_zero(zlen, to); 897 iocb->ki_pos += ret; 898 read += ret; 899 } 900 __free_pages(page, 0); 901 return read; 902 } 903 904 /* hit EOF or hole? */ 905 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && 906 ret < len) { 907 dout("sync_read hit hole, ppos %lld < size %lld" 908 ", reading more\n", iocb->ki_pos, 909 inode->i_size); 910 911 read += ret; 912 len -= ret; 913 retry_op = 0; 914 goto again; 915 } 916 } 917 918 if (ret >= 0) 919 ret += read; 920 921 return ret; 922 } 923 924 /* 925 * Take cap references to avoid releasing caps to MDS mid-write. 926 * 927 * If we are synchronous, and write with an old snap context, the OSD 928 * may return EOLDSNAPC. In that case, retry the write.. _after_ 929 * dropping our cap refs and allowing the pending snap to logically 930 * complete _before_ this write occurs. 931 * 932 * If we are near ENOSPC, write synchronously. 933 */ 934 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) 935 { 936 struct file *file = iocb->ki_filp; 937 struct ceph_file_info *fi = file->private_data; 938 struct inode *inode = file_inode(file); 939 struct ceph_inode_info *ci = ceph_inode(inode); 940 struct ceph_osd_client *osdc = 941 &ceph_sb_to_client(inode->i_sb)->client->osdc; 942 struct ceph_cap_flush *prealloc_cf; 943 ssize_t count, written = 0; 944 int err, want, got; 945 loff_t pos; 946 947 if (ceph_snap(inode) != CEPH_NOSNAP) 948 return -EROFS; 949 950 prealloc_cf = ceph_alloc_cap_flush(); 951 if (!prealloc_cf) 952 return -ENOMEM; 953 954 mutex_lock(&inode->i_mutex); 955 956 /* We can write back this queue in page reclaim */ 957 current->backing_dev_info = inode_to_bdi(inode); 958 959 err = generic_write_checks(iocb, from); 960 if (err <= 0) 961 goto out; 962 963 pos = iocb->ki_pos; 964 count = iov_iter_count(from); 965 err = file_remove_privs(file); 966 if (err) 967 goto out; 968 969 err = file_update_time(file); 970 if (err) 971 goto out; 972 973 if (ci->i_inline_version != CEPH_INLINE_NONE) { 974 err = ceph_uninline_data(file, NULL); 975 if (err < 0) 976 goto out; 977 } 978 979 retry_snap: 980 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { 981 err = -ENOSPC; 982 goto out; 983 } 984 985 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", 986 inode, ceph_vinop(inode), pos, count, inode->i_size); 987 if (fi->fmode & CEPH_FILE_MODE_LAZY) 988 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 989 else 990 want = CEPH_CAP_FILE_BUFFER; 991 got = 0; 992 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, 993 &got, NULL); 994 if (err < 0) 995 goto out; 996 997 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", 998 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); 999 1000 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || 1001 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { 1002 struct ceph_snap_context *snapc; 1003 struct iov_iter data; 1004 mutex_unlock(&inode->i_mutex); 1005 1006 spin_lock(&ci->i_ceph_lock); 1007 if (__ceph_have_pending_cap_snap(ci)) { 1008 struct ceph_cap_snap *capsnap = 1009 list_last_entry(&ci->i_cap_snaps, 1010 struct ceph_cap_snap, 1011 ci_item); 1012 snapc = ceph_get_snap_context(capsnap->context); 1013 } else { 1014 BUG_ON(!ci->i_head_snapc); 1015 snapc = ceph_get_snap_context(ci->i_head_snapc); 1016 } 1017 spin_unlock(&ci->i_ceph_lock); 1018 1019 /* we might need to revert back to that point */ 1020 data = *from; 1021 if (iocb->ki_flags & IOCB_DIRECT) 1022 written = ceph_sync_direct_write(iocb, &data, pos, 1023 snapc); 1024 else 1025 written = ceph_sync_write(iocb, &data, pos, snapc); 1026 if (written == -EOLDSNAPC) { 1027 dout("aio_write %p %llx.%llx %llu~%u" 1028 "got EOLDSNAPC, retrying\n", 1029 inode, ceph_vinop(inode), 1030 pos, (unsigned)count); 1031 mutex_lock(&inode->i_mutex); 1032 goto retry_snap; 1033 } 1034 if (written > 0) 1035 iov_iter_advance(from, written); 1036 ceph_put_snap_context(snapc); 1037 } else { 1038 loff_t old_size = inode->i_size; 1039 /* 1040 * No need to acquire the i_truncate_mutex. Because 1041 * the MDS revokes Fwb caps before sending truncate 1042 * message to us. We can't get Fwb cap while there 1043 * are pending vmtruncate. So write and vmtruncate 1044 * can not run at the same time 1045 */ 1046 written = generic_perform_write(file, from, pos); 1047 if (likely(written >= 0)) 1048 iocb->ki_pos = pos + written; 1049 if (inode->i_size > old_size) 1050 ceph_fscache_update_objectsize(inode); 1051 mutex_unlock(&inode->i_mutex); 1052 } 1053 1054 if (written >= 0) { 1055 int dirty; 1056 spin_lock(&ci->i_ceph_lock); 1057 ci->i_inline_version = CEPH_INLINE_NONE; 1058 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1059 &prealloc_cf); 1060 spin_unlock(&ci->i_ceph_lock); 1061 if (dirty) 1062 __mark_inode_dirty(inode, dirty); 1063 } 1064 1065 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", 1066 inode, ceph_vinop(inode), pos, (unsigned)count, 1067 ceph_cap_string(got)); 1068 ceph_put_cap_refs(ci, got); 1069 1070 if (written >= 0 && 1071 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || 1072 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { 1073 err = vfs_fsync_range(file, pos, pos + written - 1, 1); 1074 if (err < 0) 1075 written = err; 1076 } 1077 1078 goto out_unlocked; 1079 1080 out: 1081 mutex_unlock(&inode->i_mutex); 1082 out_unlocked: 1083 ceph_free_cap_flush(prealloc_cf); 1084 current->backing_dev_info = NULL; 1085 return written ? written : err; 1086 } 1087 1088 /* 1089 * llseek. be sure to verify file size on SEEK_END. 1090 */ 1091 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) 1092 { 1093 struct inode *inode = file->f_mapping->host; 1094 int ret; 1095 1096 mutex_lock(&inode->i_mutex); 1097 1098 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { 1099 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); 1100 if (ret < 0) { 1101 offset = ret; 1102 goto out; 1103 } 1104 } 1105 1106 switch (whence) { 1107 case SEEK_END: 1108 offset += inode->i_size; 1109 break; 1110 case SEEK_CUR: 1111 /* 1112 * Here we special-case the lseek(fd, 0, SEEK_CUR) 1113 * position-querying operation. Avoid rewriting the "same" 1114 * f_pos value back to the file because a concurrent read(), 1115 * write() or lseek() might have altered it 1116 */ 1117 if (offset == 0) { 1118 offset = file->f_pos; 1119 goto out; 1120 } 1121 offset += file->f_pos; 1122 break; 1123 case SEEK_DATA: 1124 if (offset >= inode->i_size) { 1125 ret = -ENXIO; 1126 goto out; 1127 } 1128 break; 1129 case SEEK_HOLE: 1130 if (offset >= inode->i_size) { 1131 ret = -ENXIO; 1132 goto out; 1133 } 1134 offset = inode->i_size; 1135 break; 1136 } 1137 1138 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); 1139 1140 out: 1141 mutex_unlock(&inode->i_mutex); 1142 return offset; 1143 } 1144 1145 static inline void ceph_zero_partial_page( 1146 struct inode *inode, loff_t offset, unsigned size) 1147 { 1148 struct page *page; 1149 pgoff_t index = offset >> PAGE_CACHE_SHIFT; 1150 1151 page = find_lock_page(inode->i_mapping, index); 1152 if (page) { 1153 wait_on_page_writeback(page); 1154 zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); 1155 unlock_page(page); 1156 page_cache_release(page); 1157 } 1158 } 1159 1160 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, 1161 loff_t length) 1162 { 1163 loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); 1164 if (offset < nearly) { 1165 loff_t size = nearly - offset; 1166 if (length < size) 1167 size = length; 1168 ceph_zero_partial_page(inode, offset, size); 1169 offset += size; 1170 length -= size; 1171 } 1172 if (length >= PAGE_CACHE_SIZE) { 1173 loff_t size = round_down(length, PAGE_CACHE_SIZE); 1174 truncate_pagecache_range(inode, offset, offset + size - 1); 1175 offset += size; 1176 length -= size; 1177 } 1178 if (length) 1179 ceph_zero_partial_page(inode, offset, length); 1180 } 1181 1182 static int ceph_zero_partial_object(struct inode *inode, 1183 loff_t offset, loff_t *length) 1184 { 1185 struct ceph_inode_info *ci = ceph_inode(inode); 1186 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1187 struct ceph_osd_request *req; 1188 int ret = 0; 1189 loff_t zero = 0; 1190 int op; 1191 1192 if (!length) { 1193 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; 1194 length = &zero; 1195 } else { 1196 op = CEPH_OSD_OP_ZERO; 1197 } 1198 1199 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1200 ceph_vino(inode), 1201 offset, length, 1202 0, 1, op, 1203 CEPH_OSD_FLAG_WRITE | 1204 CEPH_OSD_FLAG_ONDISK, 1205 NULL, 0, 0, false); 1206 if (IS_ERR(req)) { 1207 ret = PTR_ERR(req); 1208 goto out; 1209 } 1210 1211 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, 1212 &inode->i_mtime); 1213 1214 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1215 if (!ret) { 1216 ret = ceph_osdc_wait_request(&fsc->client->osdc, req); 1217 if (ret == -ENOENT) 1218 ret = 0; 1219 } 1220 ceph_osdc_put_request(req); 1221 1222 out: 1223 return ret; 1224 } 1225 1226 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) 1227 { 1228 int ret = 0; 1229 struct ceph_inode_info *ci = ceph_inode(inode); 1230 s32 stripe_unit = ceph_file_layout_su(ci->i_layout); 1231 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); 1232 s32 object_size = ceph_file_layout_object_size(ci->i_layout); 1233 u64 object_set_size = object_size * stripe_count; 1234 u64 nearly, t; 1235 1236 /* round offset up to next period boundary */ 1237 nearly = offset + object_set_size - 1; 1238 t = nearly; 1239 nearly -= do_div(t, object_set_size); 1240 1241 while (length && offset < nearly) { 1242 loff_t size = length; 1243 ret = ceph_zero_partial_object(inode, offset, &size); 1244 if (ret < 0) 1245 return ret; 1246 offset += size; 1247 length -= size; 1248 } 1249 while (length >= object_set_size) { 1250 int i; 1251 loff_t pos = offset; 1252 for (i = 0; i < stripe_count; ++i) { 1253 ret = ceph_zero_partial_object(inode, pos, NULL); 1254 if (ret < 0) 1255 return ret; 1256 pos += stripe_unit; 1257 } 1258 offset += object_set_size; 1259 length -= object_set_size; 1260 } 1261 while (length) { 1262 loff_t size = length; 1263 ret = ceph_zero_partial_object(inode, offset, &size); 1264 if (ret < 0) 1265 return ret; 1266 offset += size; 1267 length -= size; 1268 } 1269 return ret; 1270 } 1271 1272 static long ceph_fallocate(struct file *file, int mode, 1273 loff_t offset, loff_t length) 1274 { 1275 struct ceph_file_info *fi = file->private_data; 1276 struct inode *inode = file_inode(file); 1277 struct ceph_inode_info *ci = ceph_inode(inode); 1278 struct ceph_osd_client *osdc = 1279 &ceph_inode_to_client(inode)->client->osdc; 1280 struct ceph_cap_flush *prealloc_cf; 1281 int want, got = 0; 1282 int dirty; 1283 int ret = 0; 1284 loff_t endoff = 0; 1285 loff_t size; 1286 1287 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 1288 return -EOPNOTSUPP; 1289 1290 if (!S_ISREG(inode->i_mode)) 1291 return -EOPNOTSUPP; 1292 1293 prealloc_cf = ceph_alloc_cap_flush(); 1294 if (!prealloc_cf) 1295 return -ENOMEM; 1296 1297 mutex_lock(&inode->i_mutex); 1298 1299 if (ceph_snap(inode) != CEPH_NOSNAP) { 1300 ret = -EROFS; 1301 goto unlock; 1302 } 1303 1304 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && 1305 !(mode & FALLOC_FL_PUNCH_HOLE)) { 1306 ret = -ENOSPC; 1307 goto unlock; 1308 } 1309 1310 if (ci->i_inline_version != CEPH_INLINE_NONE) { 1311 ret = ceph_uninline_data(file, NULL); 1312 if (ret < 0) 1313 goto unlock; 1314 } 1315 1316 size = i_size_read(inode); 1317 if (!(mode & FALLOC_FL_KEEP_SIZE)) 1318 endoff = offset + length; 1319 1320 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1321 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 1322 else 1323 want = CEPH_CAP_FILE_BUFFER; 1324 1325 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); 1326 if (ret < 0) 1327 goto unlock; 1328 1329 if (mode & FALLOC_FL_PUNCH_HOLE) { 1330 if (offset < size) 1331 ceph_zero_pagecache_range(inode, offset, length); 1332 ret = ceph_zero_objects(inode, offset, length); 1333 } else if (endoff > size) { 1334 truncate_pagecache_range(inode, size, -1); 1335 if (ceph_inode_set_size(inode, endoff)) 1336 ceph_check_caps(ceph_inode(inode), 1337 CHECK_CAPS_AUTHONLY, NULL); 1338 } 1339 1340 if (!ret) { 1341 spin_lock(&ci->i_ceph_lock); 1342 ci->i_inline_version = CEPH_INLINE_NONE; 1343 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1344 &prealloc_cf); 1345 spin_unlock(&ci->i_ceph_lock); 1346 if (dirty) 1347 __mark_inode_dirty(inode, dirty); 1348 } 1349 1350 ceph_put_cap_refs(ci, got); 1351 unlock: 1352 mutex_unlock(&inode->i_mutex); 1353 ceph_free_cap_flush(prealloc_cf); 1354 return ret; 1355 } 1356 1357 const struct file_operations ceph_file_fops = { 1358 .open = ceph_open, 1359 .release = ceph_release, 1360 .llseek = ceph_llseek, 1361 .read_iter = ceph_read_iter, 1362 .write_iter = ceph_write_iter, 1363 .mmap = ceph_mmap, 1364 .fsync = ceph_fsync, 1365 .lock = ceph_lock, 1366 .flock = ceph_flock, 1367 .splice_read = generic_file_splice_read, 1368 .splice_write = iter_file_splice_write, 1369 .unlocked_ioctl = ceph_ioctl, 1370 .compat_ioctl = ceph_ioctl, 1371 .fallocate = ceph_fallocate, 1372 }; 1373 1374