1/* 2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README 3 */ 4 5#include <linux/time.h> 6#include <linux/reiserfs_fs.h> 7#include <linux/reiserfs_acl.h> 8#include <linux/reiserfs_xattr.h> 9#include <asm/uaccess.h> 10#include <linux/pagemap.h> 11#include <linux/swap.h> 12#include <linux/writeback.h> 13#include <linux/blkdev.h> 14#include <linux/buffer_head.h> 15#include <linux/quotaops.h> 16 17/* 18** We pack the tails of files on file close, not at the time they are written. 19** This implies an unnecessary copy of the tail and an unnecessary indirect item 20** insertion/balancing, for files that are written in one write. 21** It avoids unnecessary tail packings (balances) for files that are written in 22** multiple writes and are small enough to have tails. 23** 24** file_release is called by the VFS layer when the file is closed. If 25** this is the last open file descriptor, and the file 26** small enough to have a tail, and the tail is currently in an 27** unformatted node, the tail is converted back into a direct item. 28** 29** We use reiserfs_truncate_file to pack the tail, since it already has 30** all the conditions coded. 31*/ 32static int reiserfs_file_release(struct inode *inode, struct file *filp) 33{ 34 35 struct reiserfs_transaction_handle th; 36 int err; 37 int jbegin_failure = 0; 38 39 BUG_ON(!S_ISREG(inode->i_mode)); 40 41 /* fast out for when nothing needs to be done */ 42 if ((atomic_read(&inode->i_count) > 1 || 43 !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) || 44 !tail_has_to_be_packed(inode)) && 45 REISERFS_I(inode)->i_prealloc_count <= 0) { 46 return 0; 47 } 48 49 mutex_lock(&inode->i_mutex); 50 51 mutex_lock(&(REISERFS_I(inode)->i_mmap)); 52 if (REISERFS_I(inode)->i_flags & i_ever_mapped) 53 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask; 54 55 reiserfs_write_lock(inode->i_sb); 56 /* freeing preallocation only involves relogging blocks that 57 * are already in the current transaction. preallocation gets 58 * freed at the end of each transaction, so it is impossible for 59 * us to log any additional blocks (including quota blocks) 60 */ 61 err = journal_begin(&th, inode->i_sb, 1); 62 if (err) { 63 /* uh oh, we can't allow the inode to go away while there 64 * is still preallocation blocks pending. Try to join the 65 * aborted transaction 66 */ 67 jbegin_failure = err; 68 err = journal_join_abort(&th, inode->i_sb, 1); 69 70 if (err) { 71 /* hmpf, our choices here aren't good. We can pin the inode 72 * which will disallow unmount from every happening, we can 73 * do nothing, which will corrupt random memory on unmount, 74 * or we can forcibly remove the file from the preallocation 75 * list, which will leak blocks on disk. Lets pin the inode 76 * and let the admin know what is going on. 77 */ 78 igrab(inode); 79 reiserfs_warning(inode->i_sb, 80 "pinning inode %lu because the " 81 "preallocation can't be freed", 82 inode->i_ino); 83 goto out; 84 } 85 } 86 reiserfs_update_inode_transaction(inode); 87 88#ifdef REISERFS_PREALLOCATE 89 reiserfs_discard_prealloc(&th, inode); 90#endif 91 err = journal_end(&th, inode->i_sb, 1); 92 93 /* copy back the error code from journal_begin */ 94 if (!err) 95 err = jbegin_failure; 96 97 if (!err && atomic_read(&inode->i_count) <= 1 && 98 (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) && 99 tail_has_to_be_packed(inode)) { 100 /* if regular file is released by last holder and it has been 101 appended (we append by unformatted node only) or its direct 102 item(s) had to be converted, then it may have to be 103 indirect2direct converted */ 104 err = reiserfs_truncate_file(inode, 0); 105 } 106 out: 107 mutex_unlock(&(REISERFS_I(inode)->i_mmap)); 108 mutex_unlock(&inode->i_mutex); 109 reiserfs_write_unlock(inode->i_sb); 110 return err; 111} 112 113static int reiserfs_file_mmap(struct file *file, struct vm_area_struct *vma) 114{ 115 struct inode *inode; 116 117 inode = file->f_path.dentry->d_inode; 118 mutex_lock(&(REISERFS_I(inode)->i_mmap)); 119 REISERFS_I(inode)->i_flags |= i_ever_mapped; 120 mutex_unlock(&(REISERFS_I(inode)->i_mmap)); 121 122 return generic_file_mmap(file, vma); 123} 124 125static void reiserfs_vfs_truncate_file(struct inode *inode) 126{ 127 reiserfs_truncate_file(inode, 1); 128} 129 130/* Sync a reiserfs file. */ 131 132 133static int reiserfs_sync_file(struct file *p_s_filp, 134 struct dentry *p_s_dentry, int datasync) 135{ 136 struct inode *p_s_inode = p_s_dentry->d_inode; 137 int n_err; 138 int barrier_done; 139 140 BUG_ON(!S_ISREG(p_s_inode->i_mode)); 141 n_err = sync_mapping_buffers(p_s_inode->i_mapping); 142 reiserfs_write_lock(p_s_inode->i_sb); 143 barrier_done = reiserfs_commit_for_inode(p_s_inode); 144 reiserfs_write_unlock(p_s_inode->i_sb); 145 if (barrier_done != 1 && reiserfs_barrier_flush(p_s_inode->i_sb)) 146 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL); 147 if (barrier_done < 0) 148 return barrier_done; 149 return (n_err < 0) ? -EIO : 0; 150} 151 152/* I really do not want to play with memory shortage right now, so 153 to simplify the code, we are not going to write more than this much pages at 154 a time. This still should considerably improve performance compared to 4k 155 at a time case. This is 32 pages of 4k size. */ 156#define REISERFS_WRITE_PAGES_AT_A_TIME (128 * 1024) / PAGE_CACHE_SIZE 157 158/* Allocates blocks for a file to fulfil write request. 159 Maps all unmapped but prepared pages from the list. 160 Updates metadata with newly allocated blocknumbers as needed */ 161static int reiserfs_allocate_blocks_for_region(struct reiserfs_transaction_handle *th, struct inode *inode, /* Inode we work with */ 162 loff_t pos, /* Writing position */ 163 int num_pages, /* number of pages write going 164 to touch */ 165 int write_bytes, /* amount of bytes to write */ 166 struct page **prepared_pages, /* array of 167 prepared pages 168 */ 169 int blocks_to_allocate /* Amount of blocks we 170 need to allocate to 171 fit the data into file 172 */ 173 ) 174{ 175 struct cpu_key key; // cpu key of item that we are going to deal with 176 struct item_head *ih; // pointer to item head that we are going to deal with 177 struct buffer_head *bh; // Buffer head that contains items that we are going to deal with 178 __le32 *item; // pointer to item we are going to deal with 179 INITIALIZE_PATH(path); // path to item, that we are going to deal with. 180 b_blocknr_t *allocated_blocks; // Pointer to a place where allocated blocknumbers would be stored. 181 reiserfs_blocknr_hint_t hint; // hint structure for block allocator. 182 size_t res; // return value of various functions that we call. 183 int curr_block; // current block used to keep track of unmapped blocks. 184 int i; // loop counter 185 int itempos; // position in item 186 unsigned int from = (pos & (PAGE_CACHE_SIZE - 1)); // writing position in 187 // first page 188 unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1; /* last modified byte offset in last page */ 189 __u64 hole_size; // amount of blocks for a file hole, if it needed to be created. 190 int modifying_this_item = 0; // Flag for items traversal code to keep track 191 // of the fact that we already prepared 192 // current block for journal 193 int will_prealloc = 0; 194 RFALSE(!blocks_to_allocate, 195 "green-9004: tried to allocate zero blocks?"); 196 197 /* only preallocate if this is a small write */ 198 if (REISERFS_I(inode)->i_prealloc_count || 199 (!(write_bytes & (inode->i_sb->s_blocksize - 1)) && 200 blocks_to_allocate < 201 REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize)) 202 will_prealloc = 203 REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize; 204 205 allocated_blocks = kmalloc((blocks_to_allocate + will_prealloc) * 206 sizeof(b_blocknr_t), GFP_NOFS); 207 if (!allocated_blocks) 208 return -ENOMEM; 209 210 /* First we compose a key to point at the writing position, we want to do 211 that outside of any locking region. */ 212 make_cpu_key(&key, inode, pos + 1, TYPE_ANY, 3 /*key length */ ); 213 214 /* If we came here, it means we absolutely need to open a transaction, 215 since we need to allocate some blocks */ 216 reiserfs_write_lock(inode->i_sb); // Journaling stuff and we need that. 217 res = journal_begin(th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1 + 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb)); // Wish I know if this number enough 218 if (res) 219 goto error_exit; 220 reiserfs_update_inode_transaction(inode); 221 222 /* Look for the in-tree position of our write, need path for block allocator */ 223 res = search_for_position_by_key(inode->i_sb, &key, &path); 224 if (res == IO_ERROR) { 225 res = -EIO; 226 goto error_exit; 227 } 228 229 /* Allocate blocks */ 230 /* First fill in "hint" structure for block allocator */ 231 hint.th = th; // transaction handle. 232 hint.path = &path; // Path, so that block allocator can determine packing locality or whatever it needs to determine. 233 hint.inode = inode; // Inode is needed by block allocator too. 234 hint.search_start = 0; // We have no hint on where to search free blocks for block allocator. 235 hint.key = key.on_disk_key; // on disk key of file. 236 hint.block = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); // Number of disk blocks this file occupies already. 237 hint.formatted_node = 0; // We are allocating blocks for unformatted node. 238 hint.preallocate = will_prealloc; 239 240 /* Call block allocator to allocate blocks */ 241 res = 242 reiserfs_allocate_blocknrs(&hint, allocated_blocks, 243 blocks_to_allocate, blocks_to_allocate); 244 if (res != CARRY_ON) { 245 if (res == NO_DISK_SPACE) { 246 /* We flush the transaction in case of no space. This way some 247 blocks might become free */ 248 SB_JOURNAL(inode->i_sb)->j_must_wait = 1; 249 res = restart_transaction(th, inode, &path); 250 if (res) 251 goto error_exit; 252 253 /* We might have scheduled, so search again */ 254 res = 255 search_for_position_by_key(inode->i_sb, &key, 256 &path); 257 if (res == IO_ERROR) { 258 res = -EIO; 259 goto error_exit; 260 } 261 262 /* update changed info for hint structure. */ 263 res = 264 reiserfs_allocate_blocknrs(&hint, allocated_blocks, 265 blocks_to_allocate, 266 blocks_to_allocate); 267 if (res != CARRY_ON) { 268 res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC; 269 pathrelse(&path); 270 goto error_exit; 271 } 272 } else { 273 res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC; 274 pathrelse(&path); 275 goto error_exit; 276 } 277 } 278#ifdef __BIG_ENDIAN 279 // Too bad, I have not found any way to convert a given region from 280 // cpu format to little endian format 281 { 282 int i; 283 for (i = 0; i < blocks_to_allocate; i++) 284 allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]); 285 } 286#endif 287 288 /* Blocks allocating well might have scheduled and tree might have changed, 289 let's search the tree again */ 290 /* find where in the tree our write should go */ 291 res = search_for_position_by_key(inode->i_sb, &key, &path); 292 if (res == IO_ERROR) { 293 res = -EIO; 294 goto error_exit_free_blocks; 295 } 296 297 bh = get_last_bh(&path); // Get a bufferhead for last element in path. 298 ih = get_ih(&path); // Get a pointer to last item head in path. 299 item = get_item(&path); // Get a pointer to last item in path 300 301 /* Let's see what we have found */ 302 if (res != POSITION_FOUND) { /* position not found, this means that we 303 might need to append file with holes 304 first */ 305 // Since we are writing past the file's end, we need to find out if 306 // there is a hole that needs to be inserted before our writing 307 // position, and how many blocks it is going to cover (we need to 308 // populate pointers to file blocks representing the hole with zeros) 309 310 { 311 int item_offset = 1; 312 /* 313 * if ih is stat data, its offset is 0 and we don't want to 314 * add 1 to pos in the hole_size calculation 315 */ 316 if (is_statdata_le_ih(ih)) 317 item_offset = 0; 318 hole_size = (pos + item_offset - 319 (le_key_k_offset 320 (get_inode_item_key_version(inode), 321 &(ih->ih_key)) + op_bytes_number(ih, 322 inode-> 323 i_sb-> 324 s_blocksize))) 325 >> inode->i_sb->s_blocksize_bits; 326 } 327 328 if (hole_size > 0) { 329 int to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize) / UNFM_P_SIZE); // How much data to insert first time. 330 /* area filled with zeroes, to supply as list of zero blocknumbers 331 We allocate it outside of loop just in case loop would spin for 332 several iterations. */ 333 char *zeros = kzalloc(to_paste * UNFM_P_SIZE, GFP_ATOMIC); // We cannot insert more than MAX_ITEM_LEN bytes anyway. 334 if (!zeros) { 335 res = -ENOMEM; 336 goto error_exit_free_blocks; 337 } 338 do { 339 to_paste = 340 min_t(__u64, hole_size, 341 MAX_ITEM_LEN(inode->i_sb-> 342 s_blocksize) / 343 UNFM_P_SIZE); 344 if (is_indirect_le_ih(ih)) { 345 /* Ok, there is existing indirect item already. Need to append it */ 346 /* Calculate position past inserted item */ 347 make_cpu_key(&key, inode, 348 le_key_k_offset 349 (get_inode_item_key_version 350 (inode), 351 &(ih->ih_key)) + 352 op_bytes_number(ih, 353 inode-> 354 i_sb-> 355 s_blocksize), 356 TYPE_INDIRECT, 3); 357 res = 358 reiserfs_paste_into_item(th, &path, 359 &key, 360 inode, 361 (char *) 362 zeros, 363 UNFM_P_SIZE 364 * 365 to_paste); 366 if (res) { 367 kfree(zeros); 368 goto error_exit_free_blocks; 369 } 370 } else if (is_statdata_le_ih(ih)) { 371 /* No existing item, create it */ 372 /* item head for new item */ 373 struct item_head ins_ih; 374 375 /* create a key for our new item */ 376 make_cpu_key(&key, inode, 1, 377 TYPE_INDIRECT, 3); 378 379 /* Create new item head for our new item */ 380 make_le_item_head(&ins_ih, &key, 381 key.version, 1, 382 TYPE_INDIRECT, 383 to_paste * 384 UNFM_P_SIZE, 385 0 /* free space */ ); 386 387 /* Find where such item should live in the tree */ 388 res = 389 search_item(inode->i_sb, &key, 390 &path); 391 if (res != ITEM_NOT_FOUND) { 392 /* item should not exist, otherwise we have error */ 393 if (res != -ENOSPC) { 394 reiserfs_warning(inode-> 395 i_sb, 396 "green-9008: search_by_key (%K) returned %d", 397 &key, 398 res); 399 } 400 res = -EIO; 401 kfree(zeros); 402 goto error_exit_free_blocks; 403 } 404 res = 405 reiserfs_insert_item(th, &path, 406 &key, &ins_ih, 407 inode, 408 (char *)zeros); 409 } else { 410 reiserfs_panic(inode->i_sb, 411 "green-9011: Unexpected key type %K\n", 412 &key); 413 } 414 if (res) { 415 kfree(zeros); 416 goto error_exit_free_blocks; 417 } 418 /* Now we want to check if transaction is too full, and if it is 419 we restart it. This will also free the path. */ 420 if (journal_transaction_should_end 421 (th, th->t_blocks_allocated)) { 422 inode->i_size = cpu_key_k_offset(&key) + 423 (to_paste << inode->i_blkbits); 424 res = 425 restart_transaction(th, inode, 426 &path); 427 if (res) { 428 pathrelse(&path); 429 kfree(zeros); 430 goto error_exit; 431 } 432 } 433 434 /* Well, need to recalculate path and stuff */ 435 set_cpu_key_k_offset(&key, 436 cpu_key_k_offset(&key) + 437 (to_paste << inode-> 438 i_blkbits)); 439 res = 440 search_for_position_by_key(inode->i_sb, 441 &key, &path); 442 if (res == IO_ERROR) { 443 res = -EIO; 444 kfree(zeros); 445 goto error_exit_free_blocks; 446 } 447 bh = get_last_bh(&path); 448 ih = get_ih(&path); 449 item = get_item(&path); 450 hole_size -= to_paste; 451 } while (hole_size); 452 kfree(zeros); 453 } 454 } 455 // Go through existing indirect items first 456 // replace all zeroes with blocknumbers from list 457 // Note that if no corresponding item was found, by previous search, 458 // it means there are no existing in-tree representation for file area 459 // we are going to overwrite, so there is nothing to scan through for holes. 460 for (curr_block = 0, itempos = path.pos_in_item; 461 curr_block < blocks_to_allocate && res == POSITION_FOUND;) { 462 retry: 463 464 if (itempos >= ih_item_len(ih) / UNFM_P_SIZE) { 465 /* We run out of data in this indirect item, let's look for another 466 one. */ 467 /* First if we are already modifying current item, log it */ 468 if (modifying_this_item) { 469 journal_mark_dirty(th, inode->i_sb, bh); 470 modifying_this_item = 0; 471 } 472 /* Then set the key to look for a new indirect item (offset of old 473 item is added to old item length */ 474 set_cpu_key_k_offset(&key, 475 le_key_k_offset 476 (get_inode_item_key_version(inode), 477 &(ih->ih_key)) + 478 op_bytes_number(ih, 479 inode->i_sb-> 480 s_blocksize)); 481 /* Search ofor position of new key in the tree. */ 482 res = 483 search_for_position_by_key(inode->i_sb, &key, 484 &path); 485 if (res == IO_ERROR) { 486 res = -EIO; 487 goto error_exit_free_blocks; 488 } 489 bh = get_last_bh(&path); 490 ih = get_ih(&path); 491 item = get_item(&path); 492 itempos = path.pos_in_item; 493 continue; // loop to check all kinds of conditions and so on. 494 } 495 /* Ok, we have correct position in item now, so let's see if it is 496 representing file hole (blocknumber is zero) and fill it if needed */ 497 if (!item[itempos]) { 498 /* Ok, a hole. Now we need to check if we already prepared this 499 block to be journaled */ 500 while (!modifying_this_item) { // loop until succeed 501 /* Well, this item is not journaled yet, so we must prepare 502 it for journal first, before we can change it */ 503 struct item_head tmp_ih; // We copy item head of found item, 504 // here to detect if fs changed under 505 // us while we were preparing for 506 // journal. 507 int fs_gen; // We store fs generation here to find if someone 508 // changes fs under our feet 509 510 copy_item_head(&tmp_ih, ih); // Remember itemhead 511 fs_gen = get_generation(inode->i_sb); // remember fs generation 512 reiserfs_prepare_for_journal(inode->i_sb, bh, 1); // Prepare a buffer within which indirect item is stored for changing. 513 if (fs_changed(fs_gen, inode->i_sb) 514 && item_moved(&tmp_ih, &path)) { 515 // Sigh, fs was changed under us, we need to look for new 516 // location of item we are working with 517 518 /* unmark prepaerd area as journaled and search for it's 519 new position */ 520 reiserfs_restore_prepared_buffer(inode-> 521 i_sb, 522 bh); 523 res = 524 search_for_position_by_key(inode-> 525 i_sb, 526 &key, 527 &path); 528 if (res == IO_ERROR) { 529 res = -EIO; 530 goto error_exit_free_blocks; 531 } 532 bh = get_last_bh(&path); 533 ih = get_ih(&path); 534 item = get_item(&path); 535 itempos = path.pos_in_item; 536 goto retry; 537 } 538 modifying_this_item = 1; 539 } 540 item[itempos] = allocated_blocks[curr_block]; // Assign new block 541 curr_block++; 542 } 543 itempos++; 544 } 545 546 if (modifying_this_item) { // We need to log last-accessed block, if it 547 // was modified, but not logged yet. 548 journal_mark_dirty(th, inode->i_sb, bh); 549 } 550 551 if (curr_block < blocks_to_allocate) { 552 // Oh, well need to append to indirect item, or to create indirect item 553 // if there weren't any 554 if (is_indirect_le_ih(ih)) { 555 // Existing indirect item - append. First calculate key for append 556 // position. We do not need to recalculate path as it should 557 // already point to correct place. 558 make_cpu_key(&key, inode, 559 le_key_k_offset(get_inode_item_key_version 560 (inode), 561 &(ih->ih_key)) + 562 op_bytes_number(ih, 563 inode->i_sb->s_blocksize), 564 TYPE_INDIRECT, 3); 565 res = 566 reiserfs_paste_into_item(th, &path, &key, inode, 567 (char *)(allocated_blocks + 568 curr_block), 569 UNFM_P_SIZE * 570 (blocks_to_allocate - 571 curr_block)); 572 if (res) { 573 goto error_exit_free_blocks; 574 } 575 } else if (is_statdata_le_ih(ih)) { 576 // Last found item was statdata. That means we need to create indirect item. 577 struct item_head ins_ih; /* itemhead for new item */ 578 579 /* create a key for our new item */ 580 make_cpu_key(&key, inode, 1, TYPE_INDIRECT, 3); // Position one, 581 // because that's 582 // where first 583 // indirect item 584 // begins 585 /* Create new item head for our new item */ 586 make_le_item_head(&ins_ih, &key, key.version, 1, 587 TYPE_INDIRECT, 588 (blocks_to_allocate - 589 curr_block) * UNFM_P_SIZE, 590 0 /* free space */ ); 591 /* Find where such item should live in the tree */ 592 res = search_item(inode->i_sb, &key, &path); 593 if (res != ITEM_NOT_FOUND) { 594 /* Well, if we have found such item already, or some error 595 occured, we need to warn user and return error */ 596 if (res != -ENOSPC) { 597 reiserfs_warning(inode->i_sb, 598 "green-9009: search_by_key (%K) " 599 "returned %d", &key, 600 res); 601 } 602 res = -EIO; 603 goto error_exit_free_blocks; 604 } 605 /* Insert item into the tree with the data as its body */ 606 res = 607 reiserfs_insert_item(th, &path, &key, &ins_ih, 608 inode, 609 (char *)(allocated_blocks + 610 curr_block)); 611 } else { 612 reiserfs_panic(inode->i_sb, 613 "green-9010: unexpected item type for key %K\n", 614 &key); 615 } 616 } 617 // the caller is responsible for closing the transaction 618 // unless we return an error, they are also responsible for logging 619 // the inode. 620 // 621 pathrelse(&path); 622 /* 623 * cleanup prellocation from previous writes 624 * if this is a partial block write 625 */ 626 if (write_bytes & (inode->i_sb->s_blocksize - 1)) 627 reiserfs_discard_prealloc(th, inode); 628 reiserfs_write_unlock(inode->i_sb); 629 630 // go through all the pages/buffers and map the buffers to newly allocated 631 // blocks (so that system knows where to write these pages later). 632 curr_block = 0; 633 for (i = 0; i < num_pages; i++) { 634 struct page *page = prepared_pages[i]; //current page 635 struct buffer_head *head = page_buffers(page); // first buffer for a page 636 int block_start, block_end; // in-page offsets for buffers. 637 638 if (!page_buffers(page)) 639 reiserfs_panic(inode->i_sb, 640 "green-9005: No buffers for prepared page???"); 641 642 /* For each buffer in page */ 643 for (bh = head, block_start = 0; bh != head || !block_start; 644 block_start = block_end, bh = bh->b_this_page) { 645 if (!bh) 646 reiserfs_panic(inode->i_sb, 647 "green-9006: Allocated but absent buffer for a page?"); 648 block_end = block_start + inode->i_sb->s_blocksize; 649 if (i == 0 && block_end <= from) 650 /* if this buffer is before requested data to map, skip it */ 651 continue; 652 if (i == num_pages - 1 && block_start >= to) 653 /* If this buffer is after requested data to map, abort 654 processing of current page */ 655 break; 656 657 if (!buffer_mapped(bh)) { // Ok, unmapped buffer, need to map it 658 map_bh(bh, inode->i_sb, 659 le32_to_cpu(allocated_blocks 660 [curr_block])); 661 curr_block++; 662 set_buffer_new(bh); 663 } 664 } 665 } 666 667 RFALSE(curr_block > blocks_to_allocate, 668 "green-9007: Used too many blocks? weird"); 669 670 kfree(allocated_blocks); 671 return 0; 672 673// Need to deal with transaction here. 674 error_exit_free_blocks: 675 pathrelse(&path); 676 // free blocks 677 for (i = 0; i < blocks_to_allocate; i++) 678 reiserfs_free_block(th, inode, le32_to_cpu(allocated_blocks[i]), 679 1); 680 681 error_exit: 682 if (th->t_trans_id) { 683 int err; 684 // update any changes we made to blk count 685 mark_inode_dirty(inode); 686 err = 687 journal_end(th, inode->i_sb, 688 JOURNAL_PER_BALANCE_CNT * 3 + 1 + 689 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb)); 690 if (err) 691 res = err; 692 } 693 reiserfs_write_unlock(inode->i_sb); 694 kfree(allocated_blocks); 695 696 return res; 697} 698 699/* Unlock pages prepared by reiserfs_prepare_file_region_for_write */ 700static void reiserfs_unprepare_pages(struct page **prepared_pages, /* list of locked pages */ 701 size_t num_pages /* amount of pages */ ) 702{ 703 int i; // loop counter 704 705 for (i = 0; i < num_pages; i++) { 706 struct page *page = prepared_pages[i]; 707 708 try_to_free_buffers(page); 709 unlock_page(page); 710 page_cache_release(page); 711 } 712} 713 714/* This function will copy data from userspace to specified pages within 715 supplied byte range */ 716static int reiserfs_copy_from_user_to_file_region(loff_t pos, /* In-file position */ 717 int num_pages, /* Number of pages affected */ 718 int write_bytes, /* Amount of bytes to write */ 719 struct page **prepared_pages, /* pointer to 720 array to 721 prepared pages 722 */ 723 const char __user * buf /* Pointer to user-supplied 724 data */ 725 ) 726{ 727 long page_fault = 0; // status of copy_from_user. 728 int i; // loop counter. 729 int offset; // offset in page 730 731 for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages; 732 i++, offset = 0) { 733 size_t count = min_t(size_t, PAGE_CACHE_SIZE - offset, write_bytes); // How much of bytes to write to this page 734 struct page *page = prepared_pages[i]; // Current page we process. 735 736 fault_in_pages_readable(buf, count); 737 738 /* Copy data from userspace to the current page */ 739 kmap(page); 740 page_fault = __copy_from_user(page_address(page) + offset, buf, count); // Copy the data. 741 /* Flush processor's dcache for this page */ 742 flush_dcache_page(page); 743 kunmap(page); 744 buf += count; 745 write_bytes -= count; 746 747 if (page_fault) 748 break; // Was there a fault? abort. 749 } 750 751 return page_fault ? -EFAULT : 0; 752} 753 754/* taken fs/buffer.c:__block_commit_write */ 755int reiserfs_commit_page(struct inode *inode, struct page *page, 756 unsigned from, unsigned to) 757{ 758 unsigned block_start, block_end; 759 int partial = 0; 760 unsigned blocksize; 761 struct buffer_head *bh, *head; 762 unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT; 763 int new; 764 int logit = reiserfs_file_data_log(inode); 765 struct super_block *s = inode->i_sb; 766 int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize; 767 struct reiserfs_transaction_handle th; 768 int ret = 0; 769 770 th.t_trans_id = 0; 771 blocksize = 1 << inode->i_blkbits; 772 773 if (logit) { 774 reiserfs_write_lock(s); 775 ret = journal_begin(&th, s, bh_per_page + 1); 776 if (ret) 777 goto drop_write_lock; 778 reiserfs_update_inode_transaction(inode); 779 } 780 for (bh = head = page_buffers(page), block_start = 0; 781 bh != head || !block_start; 782 block_start = block_end, bh = bh->b_this_page) { 783 784 new = buffer_new(bh); 785 clear_buffer_new(bh); 786 block_end = block_start + blocksize; 787 if (block_end <= from || block_start >= to) { 788 if (!buffer_uptodate(bh)) 789 partial = 1; 790 } else { 791 set_buffer_uptodate(bh); 792 if (logit) { 793 reiserfs_prepare_for_journal(s, bh, 1); 794 journal_mark_dirty(&th, s, bh); 795 } else if (!buffer_dirty(bh)) { 796 mark_buffer_dirty(bh); 797 /* do data=ordered on any page past the end 798 * of file and any buffer marked BH_New. 799 */ 800 if (reiserfs_data_ordered(inode->i_sb) && 801 (new || page->index >= i_size_index)) { 802 reiserfs_add_ordered_list(inode, bh); 803 } 804 } 805 } 806 } 807 if (logit) { 808 ret = journal_end(&th, s, bh_per_page + 1); 809 drop_write_lock: 810 reiserfs_write_unlock(s); 811 } 812 /* 813 * If this is a partial write which happened to make all buffers 814 * uptodate then we can optimize away a bogus readpage() for 815 * the next read(). Here we 'discover' whether the page went 816 * uptodate as a result of this (potentially partial) write. 817 */ 818 if (!partial) 819 SetPageUptodate(page); 820 return ret; 821} 822 823/* Submit pages for write. This was separated from actual file copying 824 because we might want to allocate block numbers in-between. 825 This function assumes that caller will adjust file size to correct value. */ 826static int reiserfs_submit_file_region_for_write(struct reiserfs_transaction_handle *th, struct inode *inode, loff_t pos, /* Writing position offset */ 827 size_t num_pages, /* Number of pages to write */ 828 size_t write_bytes, /* number of bytes to write */ 829 struct page **prepared_pages /* list of pages */ 830 ) 831{ 832 int status; // return status of block_commit_write. 833 int retval = 0; // Return value we are going to return. 834 int i; // loop counter 835 int offset; // Writing offset in page. 836 int orig_write_bytes = write_bytes; 837 int sd_update = 0; 838 839 for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages; 840 i++, offset = 0) { 841 int count = min_t(int, PAGE_CACHE_SIZE - offset, write_bytes); // How much of bytes to write to this page 842 struct page *page = prepared_pages[i]; // Current page we process. 843 844 status = 845 reiserfs_commit_page(inode, page, offset, offset + count); 846 if (status) 847 retval = status; // To not overcomplicate matters We are going to 848 // submit all the pages even if there was error. 849 // we only remember error status to report it on 850 // exit. 851 write_bytes -= count; 852 } 853 /* now that we've gotten all the ordered buffers marked dirty, 854 * we can safely update i_size and close any running transaction 855 */ 856 if (pos + orig_write_bytes > inode->i_size) { 857 inode->i_size = pos + orig_write_bytes; // Set new size 858 /* If the file have grown so much that tail packing is no 859 * longer possible, reset "need to pack" flag */ 860 if ((have_large_tails(inode->i_sb) && 861 inode->i_size > i_block_size(inode) * 4) || 862 (have_small_tails(inode->i_sb) && 863 inode->i_size > i_block_size(inode))) 864 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask; 865 else if ((have_large_tails(inode->i_sb) && 866 inode->i_size < i_block_size(inode) * 4) || 867 (have_small_tails(inode->i_sb) && 868 inode->i_size < i_block_size(inode))) 869 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask; 870 871 if (th->t_trans_id) { 872 reiserfs_write_lock(inode->i_sb); 873 // this sets the proper flags for O_SYNC to trigger a commit 874 mark_inode_dirty(inode); 875 reiserfs_write_unlock(inode->i_sb); 876 } else { 877 reiserfs_write_lock(inode->i_sb); 878 reiserfs_update_inode_transaction(inode); 879 mark_inode_dirty(inode); 880 reiserfs_write_unlock(inode->i_sb); 881 } 882 883 sd_update = 1; 884 } 885 if (th->t_trans_id) { 886 reiserfs_write_lock(inode->i_sb); 887 if (!sd_update) 888 mark_inode_dirty(inode); 889 status = journal_end(th, th->t_super, th->t_blocks_allocated); 890 if (status) 891 retval = status; 892 reiserfs_write_unlock(inode->i_sb); 893 } 894 th->t_trans_id = 0; 895 896 /* 897 * we have to unlock the pages after updating i_size, otherwise 898 * we race with writepage 899 */ 900 for (i = 0; i < num_pages; i++) { 901 struct page *page = prepared_pages[i]; 902 unlock_page(page); 903 mark_page_accessed(page); 904 page_cache_release(page); 905 } 906 return retval; 907} 908 909/* Look if passed writing region is going to touch file's tail 910 (if it is present). And if it is, convert the tail to unformatted node */ 911static int reiserfs_check_for_tail_and_convert(struct inode *inode, /* inode to deal with */ 912 loff_t pos, /* Writing position */ 913 int write_bytes /* amount of bytes to write */ 914 ) 915{ 916 INITIALIZE_PATH(path); // needed for search_for_position 917 struct cpu_key key; // Key that would represent last touched writing byte. 918 struct item_head *ih; // item header of found block; 919 int res; // Return value of various functions we call. 920 int cont_expand_offset; // We will put offset for generic_cont_expand here 921 // This can be int just because tails are created 922 // only for small files. 923 924/* this embodies a dependency on a particular tail policy */ 925 if (inode->i_size >= inode->i_sb->s_blocksize * 4) { 926 /* such a big files do not have tails, so we won't bother ourselves 927 to look for tails, simply return */ 928 return 0; 929 } 930 931 reiserfs_write_lock(inode->i_sb); 932 /* find the item containing the last byte to be written, or if 933 * writing past the end of the file then the last item of the 934 * file (and then we check its type). */ 935 make_cpu_key(&key, inode, pos + write_bytes + 1, TYPE_ANY, 936 3 /*key length */ ); 937 res = search_for_position_by_key(inode->i_sb, &key, &path); 938 if (res == IO_ERROR) { 939 reiserfs_write_unlock(inode->i_sb); 940 return -EIO; 941 } 942 ih = get_ih(&path); 943 res = 0; 944 if (is_direct_le_ih(ih)) { 945 /* Ok, closest item is file tail (tails are stored in "direct" 946 * items), so we need to unpack it. */ 947 /* To not overcomplicate matters, we just call generic_cont_expand 948 which will in turn call other stuff and finally will boil down to 949 reiserfs_get_block() that would do necessary conversion. */ 950 cont_expand_offset = 951 le_key_k_offset(get_inode_item_key_version(inode), 952 &(ih->ih_key)); 953 pathrelse(&path); 954 res = generic_cont_expand(inode, cont_expand_offset); 955 } else 956 pathrelse(&path); 957 958 reiserfs_write_unlock(inode->i_sb); 959 return res; 960} 961 962/* This function locks pages starting from @pos for @inode. 963 @num_pages pages are locked and stored in 964 @prepared_pages array. Also buffers are allocated for these pages. 965 First and last page of the region is read if it is overwritten only 966 partially. If last page did not exist before write (file hole or file 967 append), it is zeroed, then. 968 Returns number of unallocated blocks that should be allocated to cover 969 new file data.*/ 970static int reiserfs_prepare_file_region_for_write(struct inode *inode 971 /* Inode of the file */ , 972 loff_t pos, /* position in the file */ 973 size_t num_pages, /* number of pages to 974 prepare */ 975 size_t write_bytes, /* Amount of bytes to be 976 overwritten from 977 @pos */ 978 struct page **prepared_pages /* pointer to array 979 where to store 980 prepared pages */ 981 ) 982{ 983 int res = 0; // Return values of different functions we call. 984 unsigned long index = pos >> PAGE_CACHE_SHIFT; // Offset in file in pages. 985 int from = (pos & (PAGE_CACHE_SIZE - 1)); // Writing offset in first page 986 int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1; 987 /* offset of last modified byte in last 988 page */ 989 struct address_space *mapping = inode->i_mapping; // Pages are mapped here. 990 int i; // Simple counter 991 int blocks = 0; /* Return value (blocks that should be allocated) */ 992 struct buffer_head *bh, *head; // Current bufferhead and first bufferhead 993 // of a page. 994 unsigned block_start, block_end; // Starting and ending offsets of current 995 // buffer in the page. 996 struct buffer_head *wait[2], **wait_bh = wait; // Buffers for page, if 997 // Page appeared to be not up 998 // to date. Note how we have 999 // at most 2 buffers, this is 1000 // because we at most may 1001 // partially overwrite two 1002 // buffers for one page. One at // the beginning of write area 1003 // and one at the end. 1004 // Everything inthe middle gets // overwritten totally. 1005 1006 struct cpu_key key; // cpu key of item that we are going to deal with 1007 struct item_head *ih = NULL; // pointer to item head that we are going to deal with 1008 struct buffer_head *itembuf = NULL; // Buffer head that contains items that we are going to deal with 1009 INITIALIZE_PATH(path); // path to item, that we are going to deal with. 1010 __le32 *item = NULL; // pointer to item we are going to deal with 1011 int item_pos = -1; /* Position in indirect item */ 1012 1013 if (num_pages < 1) { 1014 reiserfs_warning(inode->i_sb, 1015 "green-9001: reiserfs_prepare_file_region_for_write " 1016 "called with zero number of pages to process"); 1017 return -EFAULT; 1018 } 1019 1020 /* We have 2 loops for pages. In first loop we grab and lock the pages, so 1021 that nobody would touch these until we release the pages. Then 1022 we'd start to deal with mapping buffers to blocks. */ 1023 for (i = 0; i < num_pages; i++) { 1024 prepared_pages[i] = grab_cache_page(mapping, index + i); // locks the page 1025 if (!prepared_pages[i]) { 1026 res = -ENOMEM; 1027 goto failed_page_grabbing; 1028 } 1029 if (!page_has_buffers(prepared_pages[i])) 1030 create_empty_buffers(prepared_pages[i], 1031 inode->i_sb->s_blocksize, 0); 1032 } 1033 1034 /* Let's count amount of blocks for a case where all the blocks 1035 overwritten are new (we will substract already allocated blocks later) */ 1036 if (num_pages > 2) 1037 /* These are full-overwritten pages so we count all the blocks in 1038 these pages are counted as needed to be allocated */ 1039 blocks = 1040 (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits); 1041 1042 /* count blocks needed for first page (possibly partially written) */ 1043 blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) + !!(from & (inode->i_sb->s_blocksize - 1)); /* roundup */ 1044 1045 /* Now we account for last page. If last page == first page (we 1046 overwrite only one page), we substract all the blocks past the 1047 last writing position in a page out of already calculated number 1048 of blocks */ 1049 blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT - inode->i_blkbits)) - 1050 ((PAGE_CACHE_SIZE - to) >> inode->i_blkbits); 1051 /* Note how we do not roundup here since partial blocks still 1052 should be allocated */ 1053 1054 /* Now if all the write area lies past the file end, no point in 1055 maping blocks, since there is none, so we just zero out remaining 1056 parts of first and last pages in write area (if needed) */ 1057 if ((pos & ~((loff_t) PAGE_CACHE_SIZE - 1)) > inode->i_size) { 1058 if (from != 0) /* First page needs to be partially zeroed */ 1059 zero_user_page(prepared_pages[0], 0, from, KM_USER0); 1060 1061 if (to != PAGE_CACHE_SIZE) /* Last page needs to be partially zeroed */ 1062 zero_user_page(prepared_pages[num_pages-1], to, 1063 PAGE_CACHE_SIZE - to, KM_USER0); 1064 1065 /* Since all blocks are new - use already calculated value */ 1066 return blocks; 1067 } 1068 1069 /* Well, since we write somewhere into the middle of a file, there is 1070 possibility we are writing over some already allocated blocks, so 1071 let's map these blocks and substract number of such blocks out of blocks 1072 we need to allocate (calculated above) */ 1073 /* Mask write position to start on blocksize, we do it out of the 1074 loop for performance reasons */ 1075 pos &= ~((loff_t) inode->i_sb->s_blocksize - 1); 1076 /* Set cpu key to the starting position in a file (on left block boundary) */ 1077 make_cpu_key(&key, inode, 1078 1 + ((pos) & ~((loff_t) inode->i_sb->s_blocksize - 1)), 1079 TYPE_ANY, 3 /*key length */ ); 1080 1081 reiserfs_write_lock(inode->i_sb); // We need that for at least search_by_key() 1082 for (i = 0; i < num_pages; i++) { 1083 1084 head = page_buffers(prepared_pages[i]); 1085 /* For each buffer in the page */ 1086 for (bh = head, block_start = 0; bh != head || !block_start; 1087 block_start = block_end, bh = bh->b_this_page) { 1088 if (!bh) 1089 reiserfs_panic(inode->i_sb, 1090 "green-9002: Allocated but absent buffer for a page?"); 1091 /* Find where this buffer ends */ 1092 block_end = block_start + inode->i_sb->s_blocksize; 1093 if (i == 0 && block_end <= from) 1094 /* if this buffer is before requested data to map, skip it */ 1095 continue; 1096 1097 if (i == num_pages - 1 && block_start >= to) { 1098 /* If this buffer is after requested data to map, abort 1099 processing of current page */ 1100 break; 1101 } 1102 1103 if (buffer_mapped(bh) && bh->b_blocknr != 0) { 1104 /* This is optimisation for a case where buffer is mapped 1105 and have blocknumber assigned. In case significant amount 1106 of such buffers are present, we may avoid some amount 1107 of search_by_key calls. 1108 Probably it would be possible to move parts of this code 1109 out of BKL, but I afraid that would overcomplicate code 1110 without any noticeable benefit. 1111 */ 1112 item_pos++; 1113 /* Update the key */ 1114 set_cpu_key_k_offset(&key, 1115 cpu_key_k_offset(&key) + 1116 inode->i_sb->s_blocksize); 1117 blocks--; // Decrease the amount of blocks that need to be 1118 // allocated 1119 continue; // Go to the next buffer 1120 } 1121 1122 if (!itembuf || /* if first iteration */ 1123 item_pos >= ih_item_len(ih) / UNFM_P_SIZE) { /* or if we progressed past the 1124 current unformatted_item */ 1125 /* Try to find next item */ 1126 res = 1127 search_for_position_by_key(inode->i_sb, 1128 &key, &path); 1129 /* Abort if no more items */ 1130 if (res != POSITION_FOUND) { 1131 /* make sure later loops don't use this item */ 1132 itembuf = NULL; 1133 item = NULL; 1134 break; 1135 } 1136 1137 /* Update information about current indirect item */ 1138 itembuf = get_last_bh(&path); 1139 ih = get_ih(&path); 1140 item = get_item(&path); 1141 item_pos = path.pos_in_item; 1142 1143 RFALSE(!is_indirect_le_ih(ih), 1144 "green-9003: indirect item expected"); 1145 } 1146 1147 /* See if there is some block associated with the file 1148 at that position, map the buffer to this block */ 1149 if (get_block_num(item, item_pos)) { 1150 map_bh(bh, inode->i_sb, 1151 get_block_num(item, item_pos)); 1152 blocks--; // Decrease the amount of blocks that need to be 1153 // allocated 1154 } 1155 item_pos++; 1156 /* Update the key */ 1157 set_cpu_key_k_offset(&key, 1158 cpu_key_k_offset(&key) + 1159 inode->i_sb->s_blocksize); 1160 } 1161 } 1162 pathrelse(&path); // Free the path 1163 reiserfs_write_unlock(inode->i_sb); 1164 1165 /* Now zero out unmappend buffers for the first and last pages of 1166 write area or issue read requests if page is mapped. */ 1167 /* First page, see if it is not uptodate */ 1168 if (!PageUptodate(prepared_pages[0])) { 1169 head = page_buffers(prepared_pages[0]); 1170 1171 /* For each buffer in page */ 1172 for (bh = head, block_start = 0; bh != head || !block_start; 1173 block_start = block_end, bh = bh->b_this_page) { 1174 1175 if (!bh) 1176 reiserfs_panic(inode->i_sb, 1177 "green-9002: Allocated but absent buffer for a page?"); 1178 /* Find where this buffer ends */ 1179 block_end = block_start + inode->i_sb->s_blocksize; 1180 if (block_end <= from) 1181 /* if this buffer is before requested data to map, skip it */ 1182 continue; 1183 if (block_start < from) { /* Aha, our partial buffer */ 1184 if (buffer_mapped(bh)) { /* If it is mapped, we need to 1185 issue READ request for it to 1186 not loose data */ 1187 ll_rw_block(READ, 1, &bh); 1188 *wait_bh++ = bh; 1189 } else { /* Not mapped, zero it */ 1190 zero_user_page(prepared_pages[0], 1191 block_start, 1192 from - block_start, KM_USER0); 1193 set_buffer_uptodate(bh); 1194 } 1195 } 1196 } 1197 } 1198 1199 /* Last page, see if it is not uptodate, or if the last page is past the end of the file. */ 1200 if (!PageUptodate(prepared_pages[num_pages - 1]) || 1201 ((pos + write_bytes) >> PAGE_CACHE_SHIFT) > 1202 (inode->i_size >> PAGE_CACHE_SHIFT)) { 1203 head = page_buffers(prepared_pages[num_pages - 1]); 1204 1205 /* for each buffer in page */ 1206 for (bh = head, block_start = 0; bh != head || !block_start; 1207 block_start = block_end, bh = bh->b_this_page) { 1208 1209 if (!bh) 1210 reiserfs_panic(inode->i_sb, 1211 "green-9002: Allocated but absent buffer for a page?"); 1212 /* Find where this buffer ends */ 1213 block_end = block_start + inode->i_sb->s_blocksize; 1214 if (block_start >= to) 1215 /* if this buffer is after requested data to map, skip it */ 1216 break; 1217 if (block_end > to) { /* Aha, our partial buffer */ 1218 if (buffer_mapped(bh)) { /* If it is mapped, we need to 1219 issue READ request for it to 1220 not loose data */ 1221 ll_rw_block(READ, 1, &bh); 1222 *wait_bh++ = bh; 1223 } else { /* Not mapped, zero it */ 1224 zero_user_page(prepared_pages[num_pages-1], 1225 to, block_end - to, KM_USER0); 1226 set_buffer_uptodate(bh); 1227 } 1228 } 1229 } 1230 } 1231 1232 /* Wait for read requests we made to happen, if necessary */ 1233 while (wait_bh > wait) { 1234 wait_on_buffer(*--wait_bh); 1235 if (!buffer_uptodate(*wait_bh)) { 1236 res = -EIO; 1237 goto failed_read; 1238 } 1239 } 1240 1241 return blocks; 1242 failed_page_grabbing: 1243 num_pages = i; 1244 failed_read: 1245 reiserfs_unprepare_pages(prepared_pages, num_pages); 1246 return res; 1247} 1248 1249/* Write @count bytes at position @ppos in a file indicated by @file 1250 from the buffer @buf. 1251 1252 generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want 1253 something simple that works. It is not for serious use by general purpose filesystems, excepting the one that it was 1254 written for (ext2/3). This is for several reasons: 1255 1256 * It has no understanding of any filesystem specific optimizations. 1257 1258 * It enters the filesystem repeatedly for each page that is written. 1259 1260 * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key 1261 * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time 1262 * to reiserfs which allows for fewer tree traversals. 1263 1264 * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks. 1265 1266 * Asking the block allocation code for blocks one at a time is slightly less efficient. 1267 1268 All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to 1269 use it, but we were in a hurry to make code freeze, and so it couldn't be revised then. This new code should make 1270 things right finally. 1271 1272 Future Features: providing search_by_key with hints. 1273 1274*/ 1275static ssize_t reiserfs_file_write(struct file *file, /* the file we are going to write into */ 1276 const char __user * buf, /* pointer to user supplied data 1277 (in userspace) */ 1278 size_t count, /* amount of bytes to write */ 1279 loff_t * ppos /* pointer to position in file that we start writing at. Should be updated to 1280 * new current position before returning. */ 1281 ) 1282{ 1283 size_t already_written = 0; // Number of bytes already written to the file. 1284 loff_t pos; // Current position in the file. 1285 ssize_t res; // return value of various functions that we call. 1286 int err = 0; 1287 struct inode *inode = file->f_path.dentry->d_inode; // Inode of the file that we are writing to. 1288 /* To simplify coding at this time, we store 1289 locked pages in array for now */ 1290 struct page *prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME]; 1291 struct reiserfs_transaction_handle th; 1292 th.t_trans_id = 0; 1293 1294 /* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items 1295 * lying around (most of the disk, in fact). Despite the filesystem 1296 * now being a v3.6 format, the old items still can't support large 1297 * file sizes. Catch this case here, as the rest of the VFS layer is 1298 * oblivious to the different limitations between old and new items. 1299 * reiserfs_setattr catches this for truncates. This chunk is lifted 1300 * from generic_write_checks. */ 1301 if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 && 1302 *ppos + count > MAX_NON_LFS) { 1303 if (*ppos >= MAX_NON_LFS) { 1304 send_sig(SIGXFSZ, current, 0); 1305 return -EFBIG; 1306 } 1307 if (count > MAX_NON_LFS - (unsigned long)*ppos) 1308 count = MAX_NON_LFS - (unsigned long)*ppos; 1309 } 1310 1311 if (file->f_flags & O_DIRECT) 1312 return do_sync_write(file, buf, count, ppos); 1313 1314 if (unlikely((ssize_t) count < 0)) 1315 return -EINVAL; 1316 1317 if (unlikely(!access_ok(VERIFY_READ, buf, count))) 1318 return -EFAULT; 1319 1320 mutex_lock(&inode->i_mutex); // locks the entire file for just us 1321 1322 pos = *ppos; 1323 1324 /* Check if we can write to specified region of file, file 1325 is not overly big and this kind of stuff. Adjust pos and 1326 count, if needed */ 1327 res = generic_write_checks(file, &pos, &count, 0); 1328 if (res) 1329 goto out; 1330 1331 if (count == 0) 1332 goto out; 1333 1334 res = remove_suid(file->f_path.dentry); 1335 if (res) 1336 goto out; 1337 1338 file_update_time(file); 1339 1340 // Ok, we are done with all the checks. 1341 1342 // Now we should start real work 1343 1344 /* If we are going to write past the file's packed tail or if we are going 1345 to overwrite part of the tail, we need that tail to be converted into 1346 unformatted node */ 1347 res = reiserfs_check_for_tail_and_convert(inode, pos, count); 1348 if (res) 1349 goto out; 1350 1351 while (count > 0) { 1352 /* This is the main loop in which we running until some error occures 1353 or until we write all of the data. */ 1354 size_t num_pages; /* amount of pages we are going to write this iteration */ 1355 size_t write_bytes; /* amount of bytes to write during this iteration */ 1356 size_t blocks_to_allocate; /* how much blocks we need to allocate for this iteration */ 1357 1358 /* (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos */ 1359 num_pages = !!((pos + count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial 1360 pages */ 1361 ((count + 1362 (pos & (PAGE_CACHE_SIZE - 1))) >> PAGE_CACHE_SHIFT); 1363 /* convert size to amount of 1364 pages */ 1365 reiserfs_write_lock(inode->i_sb); 1366 if (num_pages > REISERFS_WRITE_PAGES_AT_A_TIME 1367 || num_pages > reiserfs_can_fit_pages(inode->i_sb)) { 1368 /* If we were asked to write more data than we want to or if there 1369 is not that much space, then we shorten amount of data to write 1370 for this iteration. */ 1371 num_pages = 1372 min_t(size_t, REISERFS_WRITE_PAGES_AT_A_TIME, 1373 reiserfs_can_fit_pages(inode->i_sb)); 1374 /* Also we should not forget to set size in bytes accordingly */ 1375 write_bytes = (num_pages << PAGE_CACHE_SHIFT) - 1376 (pos & (PAGE_CACHE_SIZE - 1)); 1377 /* If position is not on the 1378 start of the page, we need 1379 to substract the offset 1380 within page */ 1381 } else 1382 write_bytes = count; 1383 1384 /* reserve the blocks to be allocated later, so that later on 1385 we still have the space to write the blocks to */ 1386 reiserfs_claim_blocks_to_be_allocated(inode->i_sb, 1387 num_pages << 1388 (PAGE_CACHE_SHIFT - 1389 inode->i_blkbits)); 1390 reiserfs_write_unlock(inode->i_sb); 1391 1392 if (!num_pages) { /* If we do not have enough space even for a single page... */ 1393 if (pos > 1394 inode->i_size + inode->i_sb->s_blocksize - 1395 (pos & (inode->i_sb->s_blocksize - 1))) { 1396 res = -ENOSPC; 1397 break; // In case we are writing past the end of the last file block, break. 1398 } 1399 // Otherwise we are possibly overwriting the file, so 1400 // let's set write size to be equal or less than blocksize. 1401 // This way we get it correctly for file holes. 1402 // But overwriting files on absolutelly full volumes would not 1403 // be very efficient. Well, people are not supposed to fill 1404 // 100% of disk space anyway. 1405 write_bytes = 1406 min_t(size_t, count, 1407 inode->i_sb->s_blocksize - 1408 (pos & (inode->i_sb->s_blocksize - 1))); 1409 num_pages = 1; 1410 // No blocks were claimed before, so do it now. 1411 reiserfs_claim_blocks_to_be_allocated(inode->i_sb, 1412 1 << 1413 (PAGE_CACHE_SHIFT 1414 - 1415 inode-> 1416 i_blkbits)); 1417 } 1418 1419 /* Prepare for writing into the region, read in all the 1420 partially overwritten pages, if needed. And lock the pages, 1421 so that nobody else can access these until we are done. 1422 We get number of actual blocks needed as a result. */ 1423 res = reiserfs_prepare_file_region_for_write(inode, pos, 1424 num_pages, 1425 write_bytes, 1426 prepared_pages); 1427 if (res < 0) { 1428 reiserfs_release_claimed_blocks(inode->i_sb, 1429 num_pages << 1430 (PAGE_CACHE_SHIFT - 1431 inode->i_blkbits)); 1432 break; 1433 } 1434 1435 blocks_to_allocate = res; 1436 1437 /* First we correct our estimate of how many blocks we need */ 1438 reiserfs_release_claimed_blocks(inode->i_sb, 1439 (num_pages << 1440 (PAGE_CACHE_SHIFT - 1441 inode->i_sb-> 1442 s_blocksize_bits)) - 1443 blocks_to_allocate); 1444 1445 if (blocks_to_allocate > 0) { /*We only allocate blocks if we need to */ 1446 /* Fill in all the possible holes and append the file if needed */ 1447 res = 1448 reiserfs_allocate_blocks_for_region(&th, inode, pos, 1449 num_pages, 1450 write_bytes, 1451 prepared_pages, 1452 blocks_to_allocate); 1453 } 1454 1455 /* well, we have allocated the blocks, so it is time to free 1456 the reservation we made earlier. */ 1457 reiserfs_release_claimed_blocks(inode->i_sb, 1458 blocks_to_allocate); 1459 if (res) { 1460 reiserfs_unprepare_pages(prepared_pages, num_pages); 1461 break; 1462 } 1463 1464/* NOTE that allocating blocks and filling blocks can be done in reverse order 1465 and probably we would do that just to get rid of garbage in files after a 1466 crash */ 1467 1468 /* Copy data from user-supplied buffer to file's pages */ 1469 res = 1470 reiserfs_copy_from_user_to_file_region(pos, num_pages, 1471 write_bytes, 1472 prepared_pages, buf); 1473 if (res) { 1474 reiserfs_unprepare_pages(prepared_pages, num_pages); 1475 break; 1476 } 1477 1478 /* Send the pages to disk and unlock them. */ 1479 res = 1480 reiserfs_submit_file_region_for_write(&th, inode, pos, 1481 num_pages, 1482 write_bytes, 1483 prepared_pages); 1484 if (res) 1485 break; 1486 1487 already_written += write_bytes; 1488 buf += write_bytes; 1489 *ppos = pos += write_bytes; 1490 count -= write_bytes; 1491 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages); 1492 } 1493 1494 /* this is only true on error */ 1495 if (th.t_trans_id) { 1496 reiserfs_write_lock(inode->i_sb); 1497 err = journal_end(&th, th.t_super, th.t_blocks_allocated); 1498 reiserfs_write_unlock(inode->i_sb); 1499 if (err) { 1500 res = err; 1501 goto out; 1502 } 1503 } 1504 1505 if (likely(res >= 0) && 1506 (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode)))) 1507 res = generic_osync_inode(inode, file->f_mapping, 1508 OSYNC_METADATA | OSYNC_DATA); 1509 1510 mutex_unlock(&inode->i_mutex); 1511 reiserfs_async_progress_wait(inode->i_sb); 1512 return (already_written != 0) ? already_written : res; 1513 1514 out: 1515 mutex_unlock(&inode->i_mutex); // unlock the file on exit. 1516 return res; 1517} 1518 1519const struct file_operations reiserfs_file_operations = { 1520 .read = do_sync_read, 1521 .write = reiserfs_file_write, 1522 .ioctl = reiserfs_ioctl, 1523#ifdef CONFIG_COMPAT 1524 .compat_ioctl = reiserfs_compat_ioctl, 1525#endif 1526 .mmap = reiserfs_file_mmap, 1527 .open = generic_file_open, 1528 .release = reiserfs_file_release, 1529 .fsync = reiserfs_sync_file, 1530 .sendfile = generic_file_sendfile, 1531 .aio_read = generic_file_aio_read, 1532 .aio_write = generic_file_aio_write, 1533 .splice_read = generic_file_splice_read, 1534 .splice_write = generic_file_splice_write, 1535}; 1536 1537const struct inode_operations reiserfs_file_inode_operations = { 1538 .truncate = reiserfs_vfs_truncate_file, 1539 .setattr = reiserfs_setattr, 1540 .setxattr = reiserfs_setxattr, 1541 .getxattr = reiserfs_getxattr, 1542 .listxattr = reiserfs_listxattr, 1543 .removexattr = reiserfs_removexattr, 1544 .permission = reiserfs_permission, 1545}; 1546