Root/fs/ext3/inode.c

1/*
2 * linux/fs/ext3/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
21 *
22 * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
23 */
24
25#include <linux/module.h>
26#include <linux/fs.h>
27#include <linux/time.h>
28#include <linux/ext3_jbd.h>
29#include <linux/jbd.h>
30#include <linux/highuid.h>
31#include <linux/pagemap.h>
32#include <linux/quotaops.h>
33#include <linux/string.h>
34#include <linux/buffer_head.h>
35#include <linux/writeback.h>
36#include <linux/mpage.h>
37#include <linux/uio.h>
38#include <linux/bio.h>
39#include <linux/fiemap.h>
40#include <linux/namei.h>
41#include "xattr.h"
42#include "acl.h"
43
44static int ext3_writepage_trans_blocks(struct inode *inode);
45
46/*
47 * Test whether an inode is a fast symlink.
48 */
49static int ext3_inode_is_fast_symlink(struct inode *inode)
50{
51    int ea_blocks = EXT3_I(inode)->i_file_acl ?
52        (inode->i_sb->s_blocksize >> 9) : 0;
53
54    return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
55}
56
57/*
58 * The ext3 forget function must perform a revoke if we are freeing data
59 * which has been journaled. Metadata (eg. indirect blocks) must be
60 * revoked in all cases.
61 *
62 * "bh" may be NULL: a metadata block may have been freed from memory
63 * but there may still be a record of it in the journal, and that record
64 * still needs to be revoked.
65 */
66int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
67            struct buffer_head *bh, ext3_fsblk_t blocknr)
68{
69    int err;
70
71    might_sleep();
72
73    BUFFER_TRACE(bh, "enter");
74
75    jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
76          "data mode %lx\n",
77          bh, is_metadata, inode->i_mode,
78          test_opt(inode->i_sb, DATA_FLAGS));
79
80    /* Never use the revoke function if we are doing full data
81     * journaling: there is no need to, and a V1 superblock won't
82     * support it. Otherwise, only skip the revoke on un-journaled
83     * data blocks. */
84
85    if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
86        (!is_metadata && !ext3_should_journal_data(inode))) {
87        if (bh) {
88            BUFFER_TRACE(bh, "call journal_forget");
89            return ext3_journal_forget(handle, bh);
90        }
91        return 0;
92    }
93
94    /*
95     * data!=journal && (is_metadata || should_journal_data(inode))
96     */
97    BUFFER_TRACE(bh, "call ext3_journal_revoke");
98    err = ext3_journal_revoke(handle, blocknr, bh);
99    if (err)
100        ext3_abort(inode->i_sb, __func__,
101               "error %d when attempting revoke", err);
102    BUFFER_TRACE(bh, "exit");
103    return err;
104}
105
106/*
107 * Work out how many blocks we need to proceed with the next chunk of a
108 * truncate transaction.
109 */
110static unsigned long blocks_for_truncate(struct inode *inode)
111{
112    unsigned long needed;
113
114    needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
115
116    /* Give ourselves just enough room to cope with inodes in which
117     * i_blocks is corrupt: we've seen disk corruptions in the past
118     * which resulted in random data in an inode which looked enough
119     * like a regular file for ext3 to try to delete it. Things
120     * will go a bit crazy if that happens, but at least we should
121     * try not to panic the whole kernel. */
122    if (needed < 2)
123        needed = 2;
124
125    /* But we need to bound the transaction so we don't overflow the
126     * journal. */
127    if (needed > EXT3_MAX_TRANS_DATA)
128        needed = EXT3_MAX_TRANS_DATA;
129
130    return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
131}
132
133/*
134 * Truncate transactions can be complex and absolutely huge. So we need to
135 * be able to restart the transaction at a conventient checkpoint to make
136 * sure we don't overflow the journal.
137 *
138 * start_transaction gets us a new handle for a truncate transaction,
139 * and extend_transaction tries to extend the existing one a bit. If
140 * extend fails, we need to propagate the failure up and restart the
141 * transaction in the top-level truncate loop. --sct
142 */
143static handle_t *start_transaction(struct inode *inode)
144{
145    handle_t *result;
146
147    result = ext3_journal_start(inode, blocks_for_truncate(inode));
148    if (!IS_ERR(result))
149        return result;
150
151    ext3_std_error(inode->i_sb, PTR_ERR(result));
152    return result;
153}
154
155/*
156 * Try to extend this transaction for the purposes of truncation.
157 *
158 * Returns 0 if we managed to create more room. If we can't create more
159 * room, and the transaction must be restarted we return 1.
160 */
161static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
162{
163    if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
164        return 0;
165    if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
166        return 0;
167    return 1;
168}
169
170/*
171 * Restart the transaction associated with *handle. This does a commit,
172 * so before we call here everything must be consistently dirtied against
173 * this transaction.
174 */
175static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
176{
177    int ret;
178
179    jbd_debug(2, "restarting handle %p\n", handle);
180    /*
181     * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
182     * At this moment, get_block can be called only for blocks inside
183     * i_size since page cache has been already dropped and writes are
184     * blocked by i_mutex. So we can safely drop the truncate_mutex.
185     */
186    mutex_unlock(&EXT3_I(inode)->truncate_mutex);
187    ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
188    mutex_lock(&EXT3_I(inode)->truncate_mutex);
189    return ret;
190}
191
192/*
193 * Called at inode eviction from icache
194 */
195void ext3_evict_inode (struct inode *inode)
196{
197    struct ext3_block_alloc_info *rsv;
198    handle_t *handle;
199    int want_delete = 0;
200
201    if (!inode->i_nlink && !is_bad_inode(inode)) {
202        dquot_initialize(inode);
203        want_delete = 1;
204    }
205
206    truncate_inode_pages(&inode->i_data, 0);
207
208    ext3_discard_reservation(inode);
209    rsv = EXT3_I(inode)->i_block_alloc_info;
210    EXT3_I(inode)->i_block_alloc_info = NULL;
211    if (unlikely(rsv))
212        kfree(rsv);
213
214    if (!want_delete)
215        goto no_delete;
216
217    handle = start_transaction(inode);
218    if (IS_ERR(handle)) {
219        /*
220         * If we're going to skip the normal cleanup, we still need to
221         * make sure that the in-core orphan linked list is properly
222         * cleaned up.
223         */
224        ext3_orphan_del(NULL, inode);
225        goto no_delete;
226    }
227
228    if (IS_SYNC(inode))
229        handle->h_sync = 1;
230    inode->i_size = 0;
231    if (inode->i_blocks)
232        ext3_truncate(inode);
233    /*
234     * Kill off the orphan record which ext3_truncate created.
235     * AKPM: I think this can be inside the above `if'.
236     * Note that ext3_orphan_del() has to be able to cope with the
237     * deletion of a non-existent orphan - this is because we don't
238     * know if ext3_truncate() actually created an orphan record.
239     * (Well, we could do this if we need to, but heck - it works)
240     */
241    ext3_orphan_del(handle, inode);
242    EXT3_I(inode)->i_dtime = get_seconds();
243
244    /*
245     * One subtle ordering requirement: if anything has gone wrong
246     * (transaction abort, IO errors, whatever), then we can still
247     * do these next steps (the fs will already have been marked as
248     * having errors), but we can't free the inode if the mark_dirty
249     * fails.
250     */
251    if (ext3_mark_inode_dirty(handle, inode)) {
252        /* If that failed, just dquot_drop() and be done with that */
253        dquot_drop(inode);
254        end_writeback(inode);
255    } else {
256        ext3_xattr_delete_inode(handle, inode);
257        dquot_free_inode(inode);
258        dquot_drop(inode);
259        end_writeback(inode);
260        ext3_free_inode(handle, inode);
261    }
262    ext3_journal_stop(handle);
263    return;
264no_delete:
265    end_writeback(inode);
266    dquot_drop(inode);
267}
268
269typedef struct {
270    __le32 *p;
271    __le32 key;
272    struct buffer_head *bh;
273} Indirect;
274
275static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
276{
277    p->key = *(p->p = v);
278    p->bh = bh;
279}
280
281static int verify_chain(Indirect *from, Indirect *to)
282{
283    while (from <= to && from->key == *from->p)
284        from++;
285    return (from > to);
286}
287
288/**
289 * ext3_block_to_path - parse the block number into array of offsets
290 * @inode: inode in question (we are only interested in its superblock)
291 * @i_block: block number to be parsed
292 * @offsets: array to store the offsets in
293 * @boundary: set this non-zero if the referred-to block is likely to be
294 * followed (on disk) by an indirect block.
295 *
296 * To store the locations of file's data ext3 uses a data structure common
297 * for UNIX filesystems - tree of pointers anchored in the inode, with
298 * data blocks at leaves and indirect blocks in intermediate nodes.
299 * This function translates the block number into path in that tree -
300 * return value is the path length and @offsets[n] is the offset of
301 * pointer to (n+1)th node in the nth one. If @block is out of range
302 * (negative or too large) warning is printed and zero returned.
303 *
304 * Note: function doesn't find node addresses, so no IO is needed. All
305 * we need to know is the capacity of indirect blocks (taken from the
306 * inode->i_sb).
307 */
308
309/*
310 * Portability note: the last comparison (check that we fit into triple
311 * indirect block) is spelled differently, because otherwise on an
312 * architecture with 32-bit longs and 8Kb pages we might get into trouble
313 * if our filesystem had 8Kb blocks. We might use long long, but that would
314 * kill us on x86. Oh, well, at least the sign propagation does not matter -
315 * i_block would have to be negative in the very beginning, so we would not
316 * get there at all.
317 */
318
319static int ext3_block_to_path(struct inode *inode,
320            long i_block, int offsets[4], int *boundary)
321{
322    int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
323    int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
324    const long direct_blocks = EXT3_NDIR_BLOCKS,
325        indirect_blocks = ptrs,
326        double_blocks = (1 << (ptrs_bits * 2));
327    int n = 0;
328    int final = 0;
329
330    if (i_block < 0) {
331        ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
332    } else if (i_block < direct_blocks) {
333        offsets[n++] = i_block;
334        final = direct_blocks;
335    } else if ( (i_block -= direct_blocks) < indirect_blocks) {
336        offsets[n++] = EXT3_IND_BLOCK;
337        offsets[n++] = i_block;
338        final = ptrs;
339    } else if ((i_block -= indirect_blocks) < double_blocks) {
340        offsets[n++] = EXT3_DIND_BLOCK;
341        offsets[n++] = i_block >> ptrs_bits;
342        offsets[n++] = i_block & (ptrs - 1);
343        final = ptrs;
344    } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
345        offsets[n++] = EXT3_TIND_BLOCK;
346        offsets[n++] = i_block >> (ptrs_bits * 2);
347        offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
348        offsets[n++] = i_block & (ptrs - 1);
349        final = ptrs;
350    } else {
351        ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
352    }
353    if (boundary)
354        *boundary = final - 1 - (i_block & (ptrs - 1));
355    return n;
356}
357
358/**
359 * ext3_get_branch - read the chain of indirect blocks leading to data
360 * @inode: inode in question
361 * @depth: depth of the chain (1 - direct pointer, etc.)
362 * @offsets: offsets of pointers in inode/indirect blocks
363 * @chain: place to store the result
364 * @err: here we store the error value
365 *
366 * Function fills the array of triples <key, p, bh> and returns %NULL
367 * if everything went OK or the pointer to the last filled triple
368 * (incomplete one) otherwise. Upon the return chain[i].key contains
369 * the number of (i+1)-th block in the chain (as it is stored in memory,
370 * i.e. little-endian 32-bit), chain[i].p contains the address of that
371 * number (it points into struct inode for i==0 and into the bh->b_data
372 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
373 * block for i>0 and NULL for i==0. In other words, it holds the block
374 * numbers of the chain, addresses they were taken from (and where we can
375 * verify that chain did not change) and buffer_heads hosting these
376 * numbers.
377 *
378 * Function stops when it stumbles upon zero pointer (absent block)
379 * (pointer to last triple returned, *@err == 0)
380 * or when it gets an IO error reading an indirect block
381 * (ditto, *@err == -EIO)
382 * or when it notices that chain had been changed while it was reading
383 * (ditto, *@err == -EAGAIN)
384 * or when it reads all @depth-1 indirect blocks successfully and finds
385 * the whole chain, all way to the data (returns %NULL, *err == 0).
386 */
387static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
388                 Indirect chain[4], int *err)
389{
390    struct super_block *sb = inode->i_sb;
391    Indirect *p = chain;
392    struct buffer_head *bh;
393
394    *err = 0;
395    /* i_data is not going away, no lock needed */
396    add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
397    if (!p->key)
398        goto no_block;
399    while (--depth) {
400        bh = sb_bread(sb, le32_to_cpu(p->key));
401        if (!bh)
402            goto failure;
403        /* Reader: pointers */
404        if (!verify_chain(chain, p))
405            goto changed;
406        add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
407        /* Reader: end */
408        if (!p->key)
409            goto no_block;
410    }
411    return NULL;
412
413changed:
414    brelse(bh);
415    *err = -EAGAIN;
416    goto no_block;
417failure:
418    *err = -EIO;
419no_block:
420    return p;
421}
422
423/**
424 * ext3_find_near - find a place for allocation with sufficient locality
425 * @inode: owner
426 * @ind: descriptor of indirect block.
427 *
428 * This function returns the preferred place for block allocation.
429 * It is used when heuristic for sequential allocation fails.
430 * Rules are:
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
434 * cylinder group.
435 *
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
440 *
441 * Caller must make sure that @ind is valid and will stay that way.
442 */
443static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
444{
445    struct ext3_inode_info *ei = EXT3_I(inode);
446    __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
447    __le32 *p;
448    ext3_fsblk_t bg_start;
449    ext3_grpblk_t colour;
450
451    /* Try to find previous block */
452    for (p = ind->p - 1; p >= start; p--) {
453        if (*p)
454            return le32_to_cpu(*p);
455    }
456
457    /* No such thing, so let's try location of indirect block */
458    if (ind->bh)
459        return ind->bh->b_blocknr;
460
461    /*
462     * It is going to be referred to from the inode itself? OK, just put it
463     * into the same cylinder group then.
464     */
465    bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
466    colour = (current->pid % 16) *
467            (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
468    return bg_start + colour;
469}
470
471/**
472 * ext3_find_goal - find a preferred place for allocation.
473 * @inode: owner
474 * @block: block we want
475 * @partial: pointer to the last triple within a chain
476 *
477 * Normally this function find the preferred place for block allocation,
478 * returns it.
479 */
480
481static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
482                   Indirect *partial)
483{
484    struct ext3_block_alloc_info *block_i;
485
486    block_i = EXT3_I(inode)->i_block_alloc_info;
487
488    /*
489     * try the heuristic for sequential allocation,
490     * failing that at least try to get decent locality.
491     */
492    if (block_i && (block == block_i->last_alloc_logical_block + 1)
493        && (block_i->last_alloc_physical_block != 0)) {
494        return block_i->last_alloc_physical_block + 1;
495    }
496
497    return ext3_find_near(inode, partial);
498}
499
500/**
501 * ext3_blks_to_allocate - Look up the block map and count the number
502 * of direct blocks need to be allocated for the given branch.
503 *
504 * @branch: chain of indirect blocks
505 * @k: number of blocks need for indirect blocks
506 * @blks: number of data blocks to be mapped.
507 * @blocks_to_boundary: the offset in the indirect block
508 *
509 * return the total number of blocks to be allocate, including the
510 * direct and indirect blocks.
511 */
512static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
513        int blocks_to_boundary)
514{
515    unsigned long count = 0;
516
517    /*
518     * Simple case, [t,d]Indirect block(s) has not allocated yet
519     * then it's clear blocks on that path have not allocated
520     */
521    if (k > 0) {
522        /* right now we don't handle cross boundary allocation */
523        if (blks < blocks_to_boundary + 1)
524            count += blks;
525        else
526            count += blocks_to_boundary + 1;
527        return count;
528    }
529
530    count++;
531    while (count < blks && count <= blocks_to_boundary &&
532        le32_to_cpu(*(branch[0].p + count)) == 0) {
533        count++;
534    }
535    return count;
536}
537
538/**
539 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
540 * @handle: handle for this transaction
541 * @inode: owner
542 * @goal: preferred place for allocation
543 * @indirect_blks: the number of blocks need to allocate for indirect
544 * blocks
545 * @blks: number of blocks need to allocated for direct blocks
546 * @new_blocks: on return it will store the new block numbers for
547 * the indirect blocks(if needed) and the first direct block,
548 * @err: here we store the error value
549 *
550 * return the number of direct blocks allocated
551 */
552static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
553            ext3_fsblk_t goal, int indirect_blks, int blks,
554            ext3_fsblk_t new_blocks[4], int *err)
555{
556    int target, i;
557    unsigned long count = 0;
558    int index = 0;
559    ext3_fsblk_t current_block = 0;
560    int ret = 0;
561
562    /*
563     * Here we try to allocate the requested multiple blocks at once,
564     * on a best-effort basis.
565     * To build a branch, we should allocate blocks for
566     * the indirect blocks(if not allocated yet), and at least
567     * the first direct block of this branch. That's the
568     * minimum number of blocks need to allocate(required)
569     */
570    target = blks + indirect_blks;
571
572    while (1) {
573        count = target;
574        /* allocating blocks for indirect blocks and direct blocks */
575        current_block = ext3_new_blocks(handle,inode,goal,&count,err);
576        if (*err)
577            goto failed_out;
578
579        target -= count;
580        /* allocate blocks for indirect blocks */
581        while (index < indirect_blks && count) {
582            new_blocks[index++] = current_block++;
583            count--;
584        }
585
586        if (count > 0)
587            break;
588    }
589
590    /* save the new block number for the first direct block */
591    new_blocks[index] = current_block;
592
593    /* total number of blocks allocated for direct blocks */
594    ret = count;
595    *err = 0;
596    return ret;
597failed_out:
598    for (i = 0; i <index; i++)
599        ext3_free_blocks(handle, inode, new_blocks[i], 1);
600    return ret;
601}
602
603/**
604 * ext3_alloc_branch - allocate and set up a chain of blocks.
605 * @handle: handle for this transaction
606 * @inode: owner
607 * @indirect_blks: number of allocated indirect blocks
608 * @blks: number of allocated direct blocks
609 * @goal: preferred place for allocation
610 * @offsets: offsets (in the blocks) to store the pointers to next.
611 * @branch: place to store the chain in.
612 *
613 * This function allocates blocks, zeroes out all but the last one,
614 * links them into chain and (if we are synchronous) writes them to disk.
615 * In other words, it prepares a branch that can be spliced onto the
616 * inode. It stores the information about that chain in the branch[], in
617 * the same format as ext3_get_branch() would do. We are calling it after
618 * we had read the existing part of chain and partial points to the last
619 * triple of that (one with zero ->key). Upon the exit we have the same
620 * picture as after the successful ext3_get_block(), except that in one
621 * place chain is disconnected - *branch->p is still zero (we did not
622 * set the last link), but branch->key contains the number that should
623 * be placed into *branch->p to fill that gap.
624 *
625 * If allocation fails we free all blocks we've allocated (and forget
626 * their buffer_heads) and return the error value the from failed
627 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
628 * as described above and return 0.
629 */
630static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
631            int indirect_blks, int *blks, ext3_fsblk_t goal,
632            int *offsets, Indirect *branch)
633{
634    int blocksize = inode->i_sb->s_blocksize;
635    int i, n = 0;
636    int err = 0;
637    struct buffer_head *bh;
638    int num;
639    ext3_fsblk_t new_blocks[4];
640    ext3_fsblk_t current_block;
641
642    num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
643                *blks, new_blocks, &err);
644    if (err)
645        return err;
646
647    branch[0].key = cpu_to_le32(new_blocks[0]);
648    /*
649     * metadata blocks and data blocks are allocated.
650     */
651    for (n = 1; n <= indirect_blks; n++) {
652        /*
653         * Get buffer_head for parent block, zero it out
654         * and set the pointer to new one, then send
655         * parent to disk.
656         */
657        bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
658        branch[n].bh = bh;
659        lock_buffer(bh);
660        BUFFER_TRACE(bh, "call get_create_access");
661        err = ext3_journal_get_create_access(handle, bh);
662        if (err) {
663            unlock_buffer(bh);
664            brelse(bh);
665            goto failed;
666        }
667
668        memset(bh->b_data, 0, blocksize);
669        branch[n].p = (__le32 *) bh->b_data + offsets[n];
670        branch[n].key = cpu_to_le32(new_blocks[n]);
671        *branch[n].p = branch[n].key;
672        if ( n == indirect_blks) {
673            current_block = new_blocks[n];
674            /*
675             * End of chain, update the last new metablock of
676             * the chain to point to the new allocated
677             * data blocks numbers
678             */
679            for (i=1; i < num; i++)
680                *(branch[n].p + i) = cpu_to_le32(++current_block);
681        }
682        BUFFER_TRACE(bh, "marking uptodate");
683        set_buffer_uptodate(bh);
684        unlock_buffer(bh);
685
686        BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
687        err = ext3_journal_dirty_metadata(handle, bh);
688        if (err)
689            goto failed;
690    }
691    *blks = num;
692    return err;
693failed:
694    /* Allocation failed, free what we already allocated */
695    for (i = 1; i <= n ; i++) {
696        BUFFER_TRACE(branch[i].bh, "call journal_forget");
697        ext3_journal_forget(handle, branch[i].bh);
698    }
699    for (i = 0; i <indirect_blks; i++)
700        ext3_free_blocks(handle, inode, new_blocks[i], 1);
701
702    ext3_free_blocks(handle, inode, new_blocks[i], num);
703
704    return err;
705}
706
707/**
708 * ext3_splice_branch - splice the allocated branch onto inode.
709 * @handle: handle for this transaction
710 * @inode: owner
711 * @block: (logical) number of block we are adding
712 * @where: location of missing link
713 * @num: number of indirect blocks we are adding
714 * @blks: number of direct blocks we are adding
715 *
716 * This function fills the missing link and does all housekeeping needed in
717 * inode (->i_blocks, etc.). In case of success we end up with the full
718 * chain to new block and return 0.
719 */
720static int ext3_splice_branch(handle_t *handle, struct inode *inode,
721            long block, Indirect *where, int num, int blks)
722{
723    int i;
724    int err = 0;
725    struct ext3_block_alloc_info *block_i;
726    ext3_fsblk_t current_block;
727    struct ext3_inode_info *ei = EXT3_I(inode);
728
729    block_i = ei->i_block_alloc_info;
730    /*
731     * If we're splicing into a [td]indirect block (as opposed to the
732     * inode) then we need to get write access to the [td]indirect block
733     * before the splice.
734     */
735    if (where->bh) {
736        BUFFER_TRACE(where->bh, "get_write_access");
737        err = ext3_journal_get_write_access(handle, where->bh);
738        if (err)
739            goto err_out;
740    }
741    /* That's it */
742
743    *where->p = where->key;
744
745    /*
746     * Update the host buffer_head or inode to point to more just allocated
747     * direct blocks blocks
748     */
749    if (num == 0 && blks > 1) {
750        current_block = le32_to_cpu(where->key) + 1;
751        for (i = 1; i < blks; i++)
752            *(where->p + i ) = cpu_to_le32(current_block++);
753    }
754
755    /*
756     * update the most recently allocated logical & physical block
757     * in i_block_alloc_info, to assist find the proper goal block for next
758     * allocation
759     */
760    if (block_i) {
761        block_i->last_alloc_logical_block = block + blks - 1;
762        block_i->last_alloc_physical_block =
763                le32_to_cpu(where[num].key) + blks - 1;
764    }
765
766    /* We are done with atomic stuff, now do the rest of housekeeping */
767
768    inode->i_ctime = CURRENT_TIME_SEC;
769    ext3_mark_inode_dirty(handle, inode);
770    /* ext3_mark_inode_dirty already updated i_sync_tid */
771    atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
772
773    /* had we spliced it onto indirect block? */
774    if (where->bh) {
775        /*
776         * If we spliced it onto an indirect block, we haven't
777         * altered the inode. Note however that if it is being spliced
778         * onto an indirect block at the very end of the file (the
779         * file is growing) then we *will* alter the inode to reflect
780         * the new i_size. But that is not done here - it is done in
781         * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
782         */
783        jbd_debug(5, "splicing indirect only\n");
784        BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
785        err = ext3_journal_dirty_metadata(handle, where->bh);
786        if (err)
787            goto err_out;
788    } else {
789        /*
790         * OK, we spliced it into the inode itself on a direct block.
791         * Inode was dirtied above.
792         */
793        jbd_debug(5, "splicing direct\n");
794    }
795    return err;
796
797err_out:
798    for (i = 1; i <= num; i++) {
799        BUFFER_TRACE(where[i].bh, "call journal_forget");
800        ext3_journal_forget(handle, where[i].bh);
801        ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
802    }
803    ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
804
805    return err;
806}
807
808/*
809 * Allocation strategy is simple: if we have to allocate something, we will
810 * have to go the whole way to leaf. So let's do it before attaching anything
811 * to tree, set linkage between the newborn blocks, write them if sync is
812 * required, recheck the path, free and repeat if check fails, otherwise
813 * set the last missing link (that will protect us from any truncate-generated
814 * removals - all blocks on the path are immune now) and possibly force the
815 * write on the parent block.
816 * That has a nice additional property: no special recovery from the failed
817 * allocations is needed - we simply release blocks and do not touch anything
818 * reachable from inode.
819 *
820 * `handle' can be NULL if create == 0.
821 *
822 * The BKL may not be held on entry here. Be sure to take it early.
823 * return > 0, # of blocks mapped or allocated.
824 * return = 0, if plain lookup failed.
825 * return < 0, error case.
826 */
827int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
828        sector_t iblock, unsigned long maxblocks,
829        struct buffer_head *bh_result,
830        int create)
831{
832    int err = -EIO;
833    int offsets[4];
834    Indirect chain[4];
835    Indirect *partial;
836    ext3_fsblk_t goal;
837    int indirect_blks;
838    int blocks_to_boundary = 0;
839    int depth;
840    struct ext3_inode_info *ei = EXT3_I(inode);
841    int count = 0;
842    ext3_fsblk_t first_block = 0;
843
844
845    J_ASSERT(handle != NULL || create == 0);
846    depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
847
848    if (depth == 0)
849        goto out;
850
851    partial = ext3_get_branch(inode, depth, offsets, chain, &err);
852
853    /* Simplest case - block found, no allocation needed */
854    if (!partial) {
855        first_block = le32_to_cpu(chain[depth - 1].key);
856        clear_buffer_new(bh_result);
857        count++;
858        /*map more blocks*/
859        while (count < maxblocks && count <= blocks_to_boundary) {
860            ext3_fsblk_t blk;
861
862            if (!verify_chain(chain, chain + depth - 1)) {
863                /*
864                 * Indirect block might be removed by
865                 * truncate while we were reading it.
866                 * Handling of that case: forget what we've
867                 * got now. Flag the err as EAGAIN, so it
868                 * will reread.
869                 */
870                err = -EAGAIN;
871                count = 0;
872                break;
873            }
874            blk = le32_to_cpu(*(chain[depth-1].p + count));
875
876            if (blk == first_block + count)
877                count++;
878            else
879                break;
880        }
881        if (err != -EAGAIN)
882            goto got_it;
883    }
884
885    /* Next simple case - plain lookup or failed read of indirect block */
886    if (!create || err == -EIO)
887        goto cleanup;
888
889    mutex_lock(&ei->truncate_mutex);
890
891    /*
892     * If the indirect block is missing while we are reading
893     * the chain(ext3_get_branch() returns -EAGAIN err), or
894     * if the chain has been changed after we grab the semaphore,
895     * (either because another process truncated this branch, or
896     * another get_block allocated this branch) re-grab the chain to see if
897     * the request block has been allocated or not.
898     *
899     * Since we already block the truncate/other get_block
900     * at this point, we will have the current copy of the chain when we
901     * splice the branch into the tree.
902     */
903    if (err == -EAGAIN || !verify_chain(chain, partial)) {
904        while (partial > chain) {
905            brelse(partial->bh);
906            partial--;
907        }
908        partial = ext3_get_branch(inode, depth, offsets, chain, &err);
909        if (!partial) {
910            count++;
911            mutex_unlock(&ei->truncate_mutex);
912            if (err)
913                goto cleanup;
914            clear_buffer_new(bh_result);
915            goto got_it;
916        }
917    }
918
919    /*
920     * Okay, we need to do block allocation. Lazily initialize the block
921     * allocation info here if necessary
922    */
923    if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
924        ext3_init_block_alloc_info(inode);
925
926    goal = ext3_find_goal(inode, iblock, partial);
927
928    /* the number of blocks need to allocate for [d,t]indirect blocks */
929    indirect_blks = (chain + depth) - partial - 1;
930
931    /*
932     * Next look up the indirect map to count the totoal number of
933     * direct blocks to allocate for this branch.
934     */
935    count = ext3_blks_to_allocate(partial, indirect_blks,
936                    maxblocks, blocks_to_boundary);
937    /*
938     * Block out ext3_truncate while we alter the tree
939     */
940    err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
941                offsets + (partial - chain), partial);
942
943    /*
944     * The ext3_splice_branch call will free and forget any buffers
945     * on the new chain if there is a failure, but that risks using
946     * up transaction credits, especially for bitmaps where the
947     * credits cannot be returned. Can we handle this somehow? We
948     * may need to return -EAGAIN upwards in the worst case. --sct
949     */
950    if (!err)
951        err = ext3_splice_branch(handle, inode, iblock,
952                    partial, indirect_blks, count);
953    mutex_unlock(&ei->truncate_mutex);
954    if (err)
955        goto cleanup;
956
957    set_buffer_new(bh_result);
958got_it:
959    map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
960    if (count > blocks_to_boundary)
961        set_buffer_boundary(bh_result);
962    err = count;
963    /* Clean up and exit */
964    partial = chain + depth - 1; /* the whole chain */
965cleanup:
966    while (partial > chain) {
967        BUFFER_TRACE(partial->bh, "call brelse");
968        brelse(partial->bh);
969        partial--;
970    }
971    BUFFER_TRACE(bh_result, "returned");
972out:
973    return err;
974}
975
976/* Maximum number of blocks we map for direct IO at once. */
977#define DIO_MAX_BLOCKS 4096
978/*
979 * Number of credits we need for writing DIO_MAX_BLOCKS:
980 * We need sb + group descriptor + bitmap + inode -> 4
981 * For B blocks with A block pointers per block we need:
982 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
983 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
984 */
985#define DIO_CREDITS 25
986
987static int ext3_get_block(struct inode *inode, sector_t iblock,
988            struct buffer_head *bh_result, int create)
989{
990    handle_t *handle = ext3_journal_current_handle();
991    int ret = 0, started = 0;
992    unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
993
994    if (create && !handle) { /* Direct IO write... */
995        if (max_blocks > DIO_MAX_BLOCKS)
996            max_blocks = DIO_MAX_BLOCKS;
997        handle = ext3_journal_start(inode, DIO_CREDITS +
998                EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
999        if (IS_ERR(handle)) {
1000            ret = PTR_ERR(handle);
1001            goto out;
1002        }
1003        started = 1;
1004    }
1005
1006    ret = ext3_get_blocks_handle(handle, inode, iblock,
1007                    max_blocks, bh_result, create);
1008    if (ret > 0) {
1009        bh_result->b_size = (ret << inode->i_blkbits);
1010        ret = 0;
1011    }
1012    if (started)
1013        ext3_journal_stop(handle);
1014out:
1015    return ret;
1016}
1017
1018int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1019        u64 start, u64 len)
1020{
1021    return generic_block_fiemap(inode, fieinfo, start, len,
1022                    ext3_get_block);
1023}
1024
1025/*
1026 * `handle' can be NULL if create is zero
1027 */
1028struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1029                long block, int create, int *errp)
1030{
1031    struct buffer_head dummy;
1032    int fatal = 0, err;
1033
1034    J_ASSERT(handle != NULL || create == 0);
1035
1036    dummy.b_state = 0;
1037    dummy.b_blocknr = -1000;
1038    buffer_trace_init(&dummy.b_history);
1039    err = ext3_get_blocks_handle(handle, inode, block, 1,
1040                    &dummy, create);
1041    /*
1042     * ext3_get_blocks_handle() returns number of blocks
1043     * mapped. 0 in case of a HOLE.
1044     */
1045    if (err > 0) {
1046        if (err > 1)
1047            WARN_ON(1);
1048        err = 0;
1049    }
1050    *errp = err;
1051    if (!err && buffer_mapped(&dummy)) {
1052        struct buffer_head *bh;
1053        bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1054        if (!bh) {
1055            *errp = -EIO;
1056            goto err;
1057        }
1058        if (buffer_new(&dummy)) {
1059            J_ASSERT(create != 0);
1060            J_ASSERT(handle != NULL);
1061
1062            /*
1063             * Now that we do not always journal data, we should
1064             * keep in mind whether this should always journal the
1065             * new buffer as metadata. For now, regular file
1066             * writes use ext3_get_block instead, so it's not a
1067             * problem.
1068             */
1069            lock_buffer(bh);
1070            BUFFER_TRACE(bh, "call get_create_access");
1071            fatal = ext3_journal_get_create_access(handle, bh);
1072            if (!fatal && !buffer_uptodate(bh)) {
1073                memset(bh->b_data,0,inode->i_sb->s_blocksize);
1074                set_buffer_uptodate(bh);
1075            }
1076            unlock_buffer(bh);
1077            BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1078            err = ext3_journal_dirty_metadata(handle, bh);
1079            if (!fatal)
1080                fatal = err;
1081        } else {
1082            BUFFER_TRACE(bh, "not a new buffer");
1083        }
1084        if (fatal) {
1085            *errp = fatal;
1086            brelse(bh);
1087            bh = NULL;
1088        }
1089        return bh;
1090    }
1091err:
1092    return NULL;
1093}
1094
1095struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
1096                   int block, int create, int *err)
1097{
1098    struct buffer_head * bh;
1099
1100    bh = ext3_getblk(handle, inode, block, create, err);
1101    if (!bh)
1102        return bh;
1103    if (buffer_uptodate(bh))
1104        return bh;
1105    ll_rw_block(READ_META, 1, &bh);
1106    wait_on_buffer(bh);
1107    if (buffer_uptodate(bh))
1108        return bh;
1109    put_bh(bh);
1110    *err = -EIO;
1111    return NULL;
1112}
1113
1114static int walk_page_buffers( handle_t *handle,
1115                struct buffer_head *head,
1116                unsigned from,
1117                unsigned to,
1118                int *partial,
1119                int (*fn)( handle_t *handle,
1120                        struct buffer_head *bh))
1121{
1122    struct buffer_head *bh;
1123    unsigned block_start, block_end;
1124    unsigned blocksize = head->b_size;
1125    int err, ret = 0;
1126    struct buffer_head *next;
1127
1128    for ( bh = head, block_start = 0;
1129        ret == 0 && (bh != head || !block_start);
1130        block_start = block_end, bh = next)
1131    {
1132        next = bh->b_this_page;
1133        block_end = block_start + blocksize;
1134        if (block_end <= from || block_start >= to) {
1135            if (partial && !buffer_uptodate(bh))
1136                *partial = 1;
1137            continue;
1138        }
1139        err = (*fn)(handle, bh);
1140        if (!ret)
1141            ret = err;
1142    }
1143    return ret;
1144}
1145
1146/*
1147 * To preserve ordering, it is essential that the hole instantiation and
1148 * the data write be encapsulated in a single transaction. We cannot
1149 * close off a transaction and start a new one between the ext3_get_block()
1150 * and the commit_write(). So doing the journal_start at the start of
1151 * prepare_write() is the right place.
1152 *
1153 * Also, this function can nest inside ext3_writepage() ->
1154 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1155 * has generated enough buffer credits to do the whole page. So we won't
1156 * block on the journal in that case, which is good, because the caller may
1157 * be PF_MEMALLOC.
1158 *
1159 * By accident, ext3 can be reentered when a transaction is open via
1160 * quota file writes. If we were to commit the transaction while thus
1161 * reentered, there can be a deadlock - we would be holding a quota
1162 * lock, and the commit would never complete if another thread had a
1163 * transaction open and was blocking on the quota lock - a ranking
1164 * violation.
1165 *
1166 * So what we do is to rely on the fact that journal_stop/journal_start
1167 * will _not_ run commit under these circumstances because handle->h_ref
1168 * is elevated. We'll still have enough credits for the tiny quotafile
1169 * write.
1170 */
1171static int do_journal_get_write_access(handle_t *handle,
1172                    struct buffer_head *bh)
1173{
1174    int dirty = buffer_dirty(bh);
1175    int ret;
1176
1177    if (!buffer_mapped(bh) || buffer_freed(bh))
1178        return 0;
1179    /*
1180     * __block_prepare_write() could have dirtied some buffers. Clean
1181     * the dirty bit as jbd2_journal_get_write_access() could complain
1182     * otherwise about fs integrity issues. Setting of the dirty bit
1183     * by __block_prepare_write() isn't a real problem here as we clear
1184     * the bit before releasing a page lock and thus writeback cannot
1185     * ever write the buffer.
1186     */
1187    if (dirty)
1188        clear_buffer_dirty(bh);
1189    ret = ext3_journal_get_write_access(handle, bh);
1190    if (!ret && dirty)
1191        ret = ext3_journal_dirty_metadata(handle, bh);
1192    return ret;
1193}
1194
1195/*
1196 * Truncate blocks that were not used by write. We have to truncate the
1197 * pagecache as well so that corresponding buffers get properly unmapped.
1198 */
1199static void ext3_truncate_failed_write(struct inode *inode)
1200{
1201    truncate_inode_pages(inode->i_mapping, inode->i_size);
1202    ext3_truncate(inode);
1203}
1204
1205static int ext3_write_begin(struct file *file, struct address_space *mapping,
1206                loff_t pos, unsigned len, unsigned flags,
1207                struct page **pagep, void **fsdata)
1208{
1209    struct inode *inode = mapping->host;
1210    int ret;
1211    handle_t *handle;
1212    int retries = 0;
1213    struct page *page;
1214    pgoff_t index;
1215    unsigned from, to;
1216    /* Reserve one block more for addition to orphan list in case
1217     * we allocate blocks but write fails for some reason */
1218    int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
1219
1220    index = pos >> PAGE_CACHE_SHIFT;
1221    from = pos & (PAGE_CACHE_SIZE - 1);
1222    to = from + len;
1223
1224retry:
1225    page = grab_cache_page_write_begin(mapping, index, flags);
1226    if (!page)
1227        return -ENOMEM;
1228    *pagep = page;
1229
1230    handle = ext3_journal_start(inode, needed_blocks);
1231    if (IS_ERR(handle)) {
1232        unlock_page(page);
1233        page_cache_release(page);
1234        ret = PTR_ERR(handle);
1235        goto out;
1236    }
1237    ret = __block_write_begin(page, pos, len, ext3_get_block);
1238    if (ret)
1239        goto write_begin_failed;
1240
1241    if (ext3_should_journal_data(inode)) {
1242        ret = walk_page_buffers(handle, page_buffers(page),
1243                from, to, NULL, do_journal_get_write_access);
1244    }
1245write_begin_failed:
1246    if (ret) {
1247        /*
1248         * block_write_begin may have instantiated a few blocks
1249         * outside i_size. Trim these off again. Don't need
1250         * i_size_read because we hold i_mutex.
1251         *
1252         * Add inode to orphan list in case we crash before truncate
1253         * finishes. Do this only if ext3_can_truncate() agrees so
1254         * that orphan processing code is happy.
1255         */
1256        if (pos + len > inode->i_size && ext3_can_truncate(inode))
1257            ext3_orphan_add(handle, inode);
1258        ext3_journal_stop(handle);
1259        unlock_page(page);
1260        page_cache_release(page);
1261        if (pos + len > inode->i_size)
1262            ext3_truncate_failed_write(inode);
1263    }
1264    if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1265        goto retry;
1266out:
1267    return ret;
1268}
1269
1270
1271int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1272{
1273    int err = journal_dirty_data(handle, bh);
1274    if (err)
1275        ext3_journal_abort_handle(__func__, __func__,
1276                        bh, handle, err);
1277    return err;
1278}
1279
1280/* For ordered writepage and write_end functions */
1281static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1282{
1283    /*
1284     * Write could have mapped the buffer but it didn't copy the data in
1285     * yet. So avoid filing such buffer into a transaction.
1286     */
1287    if (buffer_mapped(bh) && buffer_uptodate(bh))
1288        return ext3_journal_dirty_data(handle, bh);
1289    return 0;
1290}
1291
1292/* For write_end() in data=journal mode */
1293static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1294{
1295    if (!buffer_mapped(bh) || buffer_freed(bh))
1296        return 0;
1297    set_buffer_uptodate(bh);
1298    return ext3_journal_dirty_metadata(handle, bh);
1299}
1300
1301/*
1302 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1303 * for the whole page but later we failed to copy the data in. Update inode
1304 * size according to what we managed to copy. The rest is going to be
1305 * truncated in write_end function.
1306 */
1307static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
1308{
1309    /* What matters to us is i_disksize. We don't write i_size anywhere */
1310    if (pos + copied > inode->i_size)
1311        i_size_write(inode, pos + copied);
1312    if (pos + copied > EXT3_I(inode)->i_disksize) {
1313        EXT3_I(inode)->i_disksize = pos + copied;
1314        mark_inode_dirty(inode);
1315    }
1316}
1317
1318/*
1319 * We need to pick up the new inode size which generic_commit_write gave us
1320 * `file' can be NULL - eg, when called from page_symlink().
1321 *
1322 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1323 * buffers are managed internally.
1324 */
1325static int ext3_ordered_write_end(struct file *file,
1326                struct address_space *mapping,
1327                loff_t pos, unsigned len, unsigned copied,
1328                struct page *page, void *fsdata)
1329{
1330    handle_t *handle = ext3_journal_current_handle();
1331    struct inode *inode = file->f_mapping->host;
1332    unsigned from, to;
1333    int ret = 0, ret2;
1334
1335    copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1336
1337    from = pos & (PAGE_CACHE_SIZE - 1);
1338    to = from + copied;
1339    ret = walk_page_buffers(handle, page_buffers(page),
1340        from, to, NULL, journal_dirty_data_fn);
1341
1342    if (ret == 0)
1343        update_file_sizes(inode, pos, copied);
1344    /*
1345     * There may be allocated blocks outside of i_size because
1346     * we failed to copy some data. Prepare for truncate.
1347     */
1348    if (pos + len > inode->i_size && ext3_can_truncate(inode))
1349        ext3_orphan_add(handle, inode);
1350    ret2 = ext3_journal_stop(handle);
1351    if (!ret)
1352        ret = ret2;
1353    unlock_page(page);
1354    page_cache_release(page);
1355
1356    if (pos + len > inode->i_size)
1357        ext3_truncate_failed_write(inode);
1358    return ret ? ret : copied;
1359}
1360
1361static int ext3_writeback_write_end(struct file *file,
1362                struct address_space *mapping,
1363                loff_t pos, unsigned len, unsigned copied,
1364                struct page *page, void *fsdata)
1365{
1366    handle_t *handle = ext3_journal_current_handle();
1367    struct inode *inode = file->f_mapping->host;
1368    int ret;
1369
1370    copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1371    update_file_sizes(inode, pos, copied);
1372    /*
1373     * There may be allocated blocks outside of i_size because
1374     * we failed to copy some data. Prepare for truncate.
1375     */
1376    if (pos + len > inode->i_size && ext3_can_truncate(inode))
1377        ext3_orphan_add(handle, inode);
1378    ret = ext3_journal_stop(handle);
1379    unlock_page(page);
1380    page_cache_release(page);
1381
1382    if (pos + len > inode->i_size)
1383        ext3_truncate_failed_write(inode);
1384    return ret ? ret : copied;
1385}
1386
1387static int ext3_journalled_write_end(struct file *file,
1388                struct address_space *mapping,
1389                loff_t pos, unsigned len, unsigned copied,
1390                struct page *page, void *fsdata)
1391{
1392    handle_t *handle = ext3_journal_current_handle();
1393    struct inode *inode = mapping->host;
1394    int ret = 0, ret2;
1395    int partial = 0;
1396    unsigned from, to;
1397
1398    from = pos & (PAGE_CACHE_SIZE - 1);
1399    to = from + len;
1400
1401    if (copied < len) {
1402        if (!PageUptodate(page))
1403            copied = 0;
1404        page_zero_new_buffers(page, from + copied, to);
1405        to = from + copied;
1406    }
1407
1408    ret = walk_page_buffers(handle, page_buffers(page), from,
1409                to, &partial, write_end_fn);
1410    if (!partial)
1411        SetPageUptodate(page);
1412
1413    if (pos + copied > inode->i_size)
1414        i_size_write(inode, pos + copied);
1415    /*
1416     * There may be allocated blocks outside of i_size because
1417     * we failed to copy some data. Prepare for truncate.
1418     */
1419    if (pos + len > inode->i_size && ext3_can_truncate(inode))
1420        ext3_orphan_add(handle, inode);
1421    ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1422    if (inode->i_size > EXT3_I(inode)->i_disksize) {
1423        EXT3_I(inode)->i_disksize = inode->i_size;
1424        ret2 = ext3_mark_inode_dirty(handle, inode);
1425        if (!ret)
1426            ret = ret2;
1427    }
1428
1429    ret2 = ext3_journal_stop(handle);
1430    if (!ret)
1431        ret = ret2;
1432    unlock_page(page);
1433    page_cache_release(page);
1434
1435    if (pos + len > inode->i_size)
1436        ext3_truncate_failed_write(inode);
1437    return ret ? ret : copied;
1438}
1439
1440/*
1441 * bmap() is special. It gets used by applications such as lilo and by
1442 * the swapper to find the on-disk block of a specific piece of data.
1443 *
1444 * Naturally, this is dangerous if the block concerned is still in the
1445 * journal. If somebody makes a swapfile on an ext3 data-journaling
1446 * filesystem and enables swap, then they may get a nasty shock when the
1447 * data getting swapped to that swapfile suddenly gets overwritten by
1448 * the original zero's written out previously to the journal and
1449 * awaiting writeback in the kernel's buffer cache.
1450 *
1451 * So, if we see any bmap calls here on a modified, data-journaled file,
1452 * take extra steps to flush any blocks which might be in the cache.
1453 */
1454static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1455{
1456    struct inode *inode = mapping->host;
1457    journal_t *journal;
1458    int err;
1459
1460    if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
1461        /*
1462         * This is a REALLY heavyweight approach, but the use of
1463         * bmap on dirty files is expected to be extremely rare:
1464         * only if we run lilo or swapon on a freshly made file
1465         * do we expect this to happen.
1466         *
1467         * (bmap requires CAP_SYS_RAWIO so this does not
1468         * represent an unprivileged user DOS attack --- we'd be
1469         * in trouble if mortal users could trigger this path at
1470         * will.)
1471         *
1472         * NB. EXT3_STATE_JDATA is not set on files other than
1473         * regular files. If somebody wants to bmap a directory
1474         * or symlink and gets confused because the buffer
1475         * hasn't yet been flushed to disk, they deserve
1476         * everything they get.
1477         */
1478
1479        ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
1480        journal = EXT3_JOURNAL(inode);
1481        journal_lock_updates(journal);
1482        err = journal_flush(journal);
1483        journal_unlock_updates(journal);
1484
1485        if (err)
1486            return 0;
1487    }
1488
1489    return generic_block_bmap(mapping,block,ext3_get_block);
1490}
1491
1492static int bget_one(handle_t *handle, struct buffer_head *bh)
1493{
1494    get_bh(bh);
1495    return 0;
1496}
1497
1498static int bput_one(handle_t *handle, struct buffer_head *bh)
1499{
1500    put_bh(bh);
1501    return 0;
1502}
1503
1504static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1505{
1506    return !buffer_mapped(bh);
1507}
1508
1509/*
1510 * Note that we always start a transaction even if we're not journalling
1511 * data. This is to preserve ordering: any hole instantiation within
1512 * __block_write_full_page -> ext3_get_block() should be journalled
1513 * along with the data so we don't crash and then get metadata which
1514 * refers to old data.
1515 *
1516 * In all journalling modes block_write_full_page() will start the I/O.
1517 *
1518 * Problem:
1519 *
1520 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1521 * ext3_writepage()
1522 *
1523 * Similar for:
1524 *
1525 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1526 *
1527 * Same applies to ext3_get_block(). We will deadlock on various things like
1528 * lock_journal and i_truncate_mutex.
1529 *
1530 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1531 * allocations fail.
1532 *
1533 * 16May01: If we're reentered then journal_current_handle() will be
1534 * non-zero. We simply *return*.
1535 *
1536 * 1 July 2001: @@@ FIXME:
1537 * In journalled data mode, a data buffer may be metadata against the
1538 * current transaction. But the same file is part of a shared mapping
1539 * and someone does a writepage() on it.
1540 *
1541 * We will move the buffer onto the async_data list, but *after* it has
1542 * been dirtied. So there's a small window where we have dirty data on
1543 * BJ_Metadata.
1544 *
1545 * Note that this only applies to the last partial page in the file. The
1546 * bit which block_write_full_page() uses prepare/commit for. (That's
1547 * broken code anyway: it's wrong for msync()).
1548 *
1549 * It's a rare case: affects the final partial page, for journalled data
1550 * where the file is subject to bith write() and writepage() in the same
1551 * transction. To fix it we'll need a custom block_write_full_page().
1552 * We'll probably need that anyway for journalling writepage() output.
1553 *
1554 * We don't honour synchronous mounts for writepage(). That would be
1555 * disastrous. Any write() or metadata operation will sync the fs for
1556 * us.
1557 *
1558 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1559 * we don't need to open a transaction here.
1560 */
1561static int ext3_ordered_writepage(struct page *page,
1562                struct writeback_control *wbc)
1563{
1564    struct inode *inode = page->mapping->host;
1565    struct buffer_head *page_bufs;
1566    handle_t *handle = NULL;
1567    int ret = 0;
1568    int err;
1569
1570    J_ASSERT(PageLocked(page));
1571    WARN_ON_ONCE(IS_RDONLY(inode));
1572
1573    /*
1574     * We give up here if we're reentered, because it might be for a
1575     * different filesystem.
1576     */
1577    if (ext3_journal_current_handle())
1578        goto out_fail;
1579
1580    if (!page_has_buffers(page)) {
1581        create_empty_buffers(page, inode->i_sb->s_blocksize,
1582                (1 << BH_Dirty)|(1 << BH_Uptodate));
1583        page_bufs = page_buffers(page);
1584    } else {
1585        page_bufs = page_buffers(page);
1586        if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1587                       NULL, buffer_unmapped)) {
1588            /* Provide NULL get_block() to catch bugs if buffers
1589             * weren't really mapped */
1590            return block_write_full_page(page, NULL, wbc);
1591        }
1592    }
1593    handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1594
1595    if (IS_ERR(handle)) {
1596        ret = PTR_ERR(handle);
1597        goto out_fail;
1598    }
1599
1600    walk_page_buffers(handle, page_bufs, 0,
1601            PAGE_CACHE_SIZE, NULL, bget_one);
1602
1603    ret = block_write_full_page(page, ext3_get_block, wbc);
1604
1605    /*
1606     * The page can become unlocked at any point now, and
1607     * truncate can then come in and change things. So we
1608     * can't touch *page from now on. But *page_bufs is
1609     * safe due to elevated refcount.
1610     */
1611
1612    /*
1613     * And attach them to the current transaction. But only if
1614     * block_write_full_page() succeeded. Otherwise they are unmapped,
1615     * and generally junk.
1616     */
1617    if (ret == 0) {
1618        err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1619                    NULL, journal_dirty_data_fn);
1620        if (!ret)
1621            ret = err;
1622    }
1623    walk_page_buffers(handle, page_bufs, 0,
1624            PAGE_CACHE_SIZE, NULL, bput_one);
1625    err = ext3_journal_stop(handle);
1626    if (!ret)
1627        ret = err;
1628    return ret;
1629
1630out_fail:
1631    redirty_page_for_writepage(wbc, page);
1632    unlock_page(page);
1633    return ret;
1634}
1635
1636static int ext3_writeback_writepage(struct page *page,
1637                struct writeback_control *wbc)
1638{
1639    struct inode *inode = page->mapping->host;
1640    handle_t *handle = NULL;
1641    int ret = 0;
1642    int err;
1643
1644    J_ASSERT(PageLocked(page));
1645    WARN_ON_ONCE(IS_RDONLY(inode));
1646
1647    if (ext3_journal_current_handle())
1648        goto out_fail;
1649
1650    if (page_has_buffers(page)) {
1651        if (!walk_page_buffers(NULL, page_buffers(page), 0,
1652                      PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1653            /* Provide NULL get_block() to catch bugs if buffers
1654             * weren't really mapped */
1655            return block_write_full_page(page, NULL, wbc);
1656        }
1657    }
1658
1659    handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1660    if (IS_ERR(handle)) {
1661        ret = PTR_ERR(handle);
1662        goto out_fail;
1663    }
1664
1665    ret = block_write_full_page(page, ext3_get_block, wbc);
1666
1667    err = ext3_journal_stop(handle);
1668    if (!ret)
1669        ret = err;
1670    return ret;
1671
1672out_fail:
1673    redirty_page_for_writepage(wbc, page);
1674    unlock_page(page);
1675    return ret;
1676}
1677
1678static int ext3_journalled_writepage(struct page *page,
1679                struct writeback_control *wbc)
1680{
1681    struct inode *inode = page->mapping->host;
1682    handle_t *handle = NULL;
1683    int ret = 0;
1684    int err;
1685
1686    J_ASSERT(PageLocked(page));
1687    WARN_ON_ONCE(IS_RDONLY(inode));
1688
1689    if (ext3_journal_current_handle())
1690        goto no_write;
1691
1692    handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1693    if (IS_ERR(handle)) {
1694        ret = PTR_ERR(handle);
1695        goto no_write;
1696    }
1697
1698    if (!page_has_buffers(page) || PageChecked(page)) {
1699        /*
1700         * It's mmapped pagecache. Add buffers and journal it. There
1701         * doesn't seem much point in redirtying the page here.
1702         */
1703        ClearPageChecked(page);
1704        ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1705                      ext3_get_block);
1706        if (ret != 0) {
1707            ext3_journal_stop(handle);
1708            goto out_unlock;
1709        }
1710        ret = walk_page_buffers(handle, page_buffers(page), 0,
1711            PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1712
1713        err = walk_page_buffers(handle, page_buffers(page), 0,
1714                PAGE_CACHE_SIZE, NULL, write_end_fn);
1715        if (ret == 0)
1716            ret = err;
1717        ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1718        unlock_page(page);
1719    } else {
1720        /*
1721         * It may be a page full of checkpoint-mode buffers. We don't
1722         * really know unless we go poke around in the buffer_heads.
1723         * But block_write_full_page will do the right thing.
1724         */
1725        ret = block_write_full_page(page, ext3_get_block, wbc);
1726    }
1727    err = ext3_journal_stop(handle);
1728    if (!ret)
1729        ret = err;
1730out:
1731    return ret;
1732
1733no_write:
1734    redirty_page_for_writepage(wbc, page);
1735out_unlock:
1736    unlock_page(page);
1737    goto out;
1738}
1739
1740static int ext3_readpage(struct file *file, struct page *page)
1741{
1742    return mpage_readpage(page, ext3_get_block);
1743}
1744
1745static int
1746ext3_readpages(struct file *file, struct address_space *mapping,
1747        struct list_head *pages, unsigned nr_pages)
1748{
1749    return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1750}
1751
1752static void ext3_invalidatepage(struct page *page, unsigned long offset)
1753{
1754    journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1755
1756    /*
1757     * If it's a full truncate we just forget about the pending dirtying
1758     */
1759    if (offset == 0)
1760        ClearPageChecked(page);
1761
1762    journal_invalidatepage(journal, page, offset);
1763}
1764
1765static int ext3_releasepage(struct page *page, gfp_t wait)
1766{
1767    journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1768
1769    WARN_ON(PageChecked(page));
1770    if (!page_has_buffers(page))
1771        return 0;
1772    return journal_try_to_free_buffers(journal, page, wait);
1773}
1774
1775/*
1776 * If the O_DIRECT write will extend the file then add this inode to the
1777 * orphan list. So recovery will truncate it back to the original size
1778 * if the machine crashes during the write.
1779 *
1780 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1781 * crashes then stale disk data _may_ be exposed inside the file. But current
1782 * VFS code falls back into buffered path in that case so we are safe.
1783 */
1784static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1785            const struct iovec *iov, loff_t offset,
1786            unsigned long nr_segs)
1787{
1788    struct file *file = iocb->ki_filp;
1789    struct inode *inode = file->f_mapping->host;
1790    struct ext3_inode_info *ei = EXT3_I(inode);
1791    handle_t *handle;
1792    ssize_t ret;
1793    int orphan = 0;
1794    size_t count = iov_length(iov, nr_segs);
1795    int retries = 0;
1796
1797    if (rw == WRITE) {
1798        loff_t final_size = offset + count;
1799
1800        if (final_size > inode->i_size) {
1801            /* Credits for sb + inode write */
1802            handle = ext3_journal_start(inode, 2);
1803            if (IS_ERR(handle)) {
1804                ret = PTR_ERR(handle);
1805                goto out;
1806            }
1807            ret = ext3_orphan_add(handle, inode);
1808            if (ret) {
1809                ext3_journal_stop(handle);
1810                goto out;
1811            }
1812            orphan = 1;
1813            ei->i_disksize = inode->i_size;
1814            ext3_journal_stop(handle);
1815        }
1816    }
1817
1818retry:
1819    ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1820                 offset, nr_segs,
1821                 ext3_get_block, NULL);
1822    /*
1823     * In case of error extending write may have instantiated a few
1824     * blocks outside i_size. Trim these off again.
1825     */
1826    if (unlikely((rw & WRITE) && ret < 0)) {
1827        loff_t isize = i_size_read(inode);
1828        loff_t end = offset + iov_length(iov, nr_segs);
1829
1830        if (end > isize)
1831            vmtruncate(inode, isize);
1832    }
1833    if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1834        goto retry;
1835
1836    if (orphan) {
1837        int err;
1838
1839        /* Credits for sb + inode write */
1840        handle = ext3_journal_start(inode, 2);
1841        if (IS_ERR(handle)) {
1842            /* This is really bad luck. We've written the data
1843             * but cannot extend i_size. Truncate allocated blocks
1844             * and pretend the write failed... */
1845            ext3_truncate(inode);
1846            ret = PTR_ERR(handle);
1847            goto out;
1848        }
1849        if (inode->i_nlink)
1850            ext3_orphan_del(handle, inode);
1851        if (ret > 0) {
1852            loff_t end = offset + ret;
1853            if (end > inode->i_size) {
1854                ei->i_disksize = end;
1855                i_size_write(inode, end);
1856                /*
1857                 * We're going to return a positive `ret'
1858                 * here due to non-zero-length I/O, so there's
1859                 * no way of reporting error returns from
1860                 * ext3_mark_inode_dirty() to userspace. So
1861                 * ignore it.
1862                 */
1863                ext3_mark_inode_dirty(handle, inode);
1864            }
1865        }
1866        err = ext3_journal_stop(handle);
1867        if (ret == 0)
1868            ret = err;
1869    }
1870out:
1871    return ret;
1872}
1873
1874/*
1875 * Pages can be marked dirty completely asynchronously from ext3's journalling
1876 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1877 * much here because ->set_page_dirty is called under VFS locks. The page is
1878 * not necessarily locked.
1879 *
1880 * We cannot just dirty the page and leave attached buffers clean, because the
1881 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1882 * or jbddirty because all the journalling code will explode.
1883 *
1884 * So what we do is to mark the page "pending dirty" and next time writepage
1885 * is called, propagate that into the buffers appropriately.
1886 */
1887static int ext3_journalled_set_page_dirty(struct page *page)
1888{
1889    SetPageChecked(page);
1890    return __set_page_dirty_nobuffers(page);
1891}
1892
1893static const struct address_space_operations ext3_ordered_aops = {
1894    .readpage = ext3_readpage,
1895    .readpages = ext3_readpages,
1896    .writepage = ext3_ordered_writepage,
1897    .write_begin = ext3_write_begin,
1898    .write_end = ext3_ordered_write_end,
1899    .bmap = ext3_bmap,
1900    .invalidatepage = ext3_invalidatepage,
1901    .releasepage = ext3_releasepage,
1902    .direct_IO = ext3_direct_IO,
1903    .migratepage = buffer_migrate_page,
1904    .is_partially_uptodate = block_is_partially_uptodate,
1905    .error_remove_page = generic_error_remove_page,
1906};
1907
1908static const struct address_space_operations ext3_writeback_aops = {
1909    .readpage = ext3_readpage,
1910    .readpages = ext3_readpages,
1911    .writepage = ext3_writeback_writepage,
1912    .write_begin = ext3_write_begin,
1913    .write_end = ext3_writeback_write_end,
1914    .bmap = ext3_bmap,
1915    .invalidatepage = ext3_invalidatepage,
1916    .releasepage = ext3_releasepage,
1917    .direct_IO = ext3_direct_IO,
1918    .migratepage = buffer_migrate_page,
1919    .is_partially_uptodate = block_is_partially_uptodate,
1920    .error_remove_page = generic_error_remove_page,
1921};
1922
1923static const struct address_space_operations ext3_journalled_aops = {
1924    .readpage = ext3_readpage,
1925    .readpages = ext3_readpages,
1926    .writepage = ext3_journalled_writepage,
1927    .write_begin = ext3_write_begin,
1928    .write_end = ext3_journalled_write_end,
1929    .set_page_dirty = ext3_journalled_set_page_dirty,
1930    .bmap = ext3_bmap,
1931    .invalidatepage = ext3_invalidatepage,
1932    .releasepage = ext3_releasepage,
1933    .is_partially_uptodate = block_is_partially_uptodate,
1934    .error_remove_page = generic_error_remove_page,
1935};
1936
1937void ext3_set_aops(struct inode *inode)
1938{
1939    if (ext3_should_order_data(inode))
1940        inode->i_mapping->a_ops = &ext3_ordered_aops;
1941    else if (ext3_should_writeback_data(inode))
1942        inode->i_mapping->a_ops = &ext3_writeback_aops;
1943    else
1944        inode->i_mapping->a_ops = &ext3_journalled_aops;
1945}
1946
1947/*
1948 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1949 * up to the end of the block which corresponds to `from'.
1950 * This required during truncate. We need to physically zero the tail end
1951 * of that block so it doesn't yield old data if the file is later grown.
1952 */
1953static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1954        struct address_space *mapping, loff_t from)
1955{
1956    ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1957    unsigned offset = from & (PAGE_CACHE_SIZE-1);
1958    unsigned blocksize, iblock, length, pos;
1959    struct inode *inode = mapping->host;
1960    struct buffer_head *bh;
1961    int err = 0;
1962
1963    blocksize = inode->i_sb->s_blocksize;
1964    length = blocksize - (offset & (blocksize - 1));
1965    iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1966
1967    if (!page_has_buffers(page))
1968        create_empty_buffers(page, blocksize, 0);
1969
1970    /* Find the buffer that contains "offset" */
1971    bh = page_buffers(page);
1972    pos = blocksize;
1973    while (offset >= pos) {
1974        bh = bh->b_this_page;
1975        iblock++;
1976        pos += blocksize;
1977    }
1978
1979    err = 0;
1980    if (buffer_freed(bh)) {
1981        BUFFER_TRACE(bh, "freed: skip");
1982        goto unlock;
1983    }
1984
1985    if (!buffer_mapped(bh)) {
1986        BUFFER_TRACE(bh, "unmapped");
1987        ext3_get_block(inode, iblock, bh, 0);
1988        /* unmapped? It's a hole - nothing to do */
1989        if (!buffer_mapped(bh)) {
1990            BUFFER_TRACE(bh, "still unmapped");
1991            goto unlock;
1992        }
1993    }
1994
1995    /* Ok, it's mapped. Make sure it's up-to-date */
1996    if (PageUptodate(page))
1997        set_buffer_uptodate(bh);
1998
1999    if (!buffer_uptodate(bh)) {
2000        err = -EIO;
2001        ll_rw_block(READ, 1, &bh);
2002        wait_on_buffer(bh);
2003        /* Uhhuh. Read error. Complain and punt. */
2004        if (!buffer_uptodate(bh))
2005            goto unlock;
2006    }
2007
2008    if (ext3_should_journal_data(inode)) {
2009        BUFFER_TRACE(bh, "get write access");
2010        err = ext3_journal_get_write_access(handle, bh);
2011        if (err)
2012            goto unlock;
2013    }
2014
2015    zero_user(page, offset, length);
2016    BUFFER_TRACE(bh, "zeroed end of block");
2017
2018    err = 0;
2019    if (ext3_should_journal_data(inode)) {
2020        err = ext3_journal_dirty_metadata(handle, bh);
2021    } else {
2022        if (ext3_should_order_data(inode))
2023            err = ext3_journal_dirty_data(handle, bh);
2024        mark_buffer_dirty(bh);
2025    }
2026
2027unlock:
2028    unlock_page(page);
2029    page_cache_release(page);
2030    return err;
2031}
2032
2033/*
2034 * Probably it should be a library function... search for first non-zero word
2035 * or memcmp with zero_page, whatever is better for particular architecture.
2036 * Linus?
2037 */
2038static inline int all_zeroes(__le32 *p, __le32 *q)
2039{
2040    while (p < q)
2041        if (*p++)
2042            return 0;
2043    return 1;
2044}
2045
2046/**
2047 * ext3_find_shared - find the indirect blocks for partial truncation.
2048 * @inode: inode in question
2049 * @depth: depth of the affected branch
2050 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2051 * @chain: place to store the pointers to partial indirect blocks
2052 * @top: place to the (detached) top of branch
2053 *
2054 * This is a helper function used by ext3_truncate().
2055 *
2056 * When we do truncate() we may have to clean the ends of several
2057 * indirect blocks but leave the blocks themselves alive. Block is
2058 * partially truncated if some data below the new i_size is referred
2059 * from it (and it is on the path to the first completely truncated
2060 * data block, indeed). We have to free the top of that path along
2061 * with everything to the right of the path. Since no allocation
2062 * past the truncation point is possible until ext3_truncate()
2063 * finishes, we may safely do the latter, but top of branch may
2064 * require special attention - pageout below the truncation point
2065 * might try to populate it.
2066 *
2067 * We atomically detach the top of branch from the tree, store the
2068 * block number of its root in *@top, pointers to buffer_heads of
2069 * partially truncated blocks - in @chain[].bh and pointers to
2070 * their last elements that should not be removed - in
2071 * @chain[].p. Return value is the pointer to last filled element
2072 * of @chain.
2073 *
2074 * The work left to caller to do the actual freeing of subtrees:
2075 * a) free the subtree starting from *@top
2076 * b) free the subtrees whose roots are stored in
2077 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2078 * c) free the subtrees growing from the inode past the @chain[0].
2079 * (no partially truncated stuff there). */
2080
2081static Indirect *ext3_find_shared(struct inode *inode, int depth,
2082            int offsets[4], Indirect chain[4], __le32 *top)
2083{
2084    Indirect *partial, *p;
2085    int k, err;
2086
2087    *top = 0;
2088    /* Make k index the deepest non-null offset + 1 */
2089    for (k = depth; k > 1 && !offsets[k-1]; k--)
2090        ;
2091    partial = ext3_get_branch(inode, k, offsets, chain, &err);
2092    /* Writer: pointers */
2093    if (!partial)
2094        partial = chain + k-1;
2095    /*
2096     * If the branch acquired continuation since we've looked at it -
2097     * fine, it should all survive and (new) top doesn't belong to us.
2098     */
2099    if (!partial->key && *partial->p)
2100        /* Writer: end */
2101        goto no_top;
2102    for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2103        ;
2104    /*
2105     * OK, we've found the last block that must survive. The rest of our
2106     * branch should be detached before unlocking. However, if that rest
2107     * of branch is all ours and does not grow immediately from the inode
2108     * it's easier to cheat and just decrement partial->p.
2109     */
2110    if (p == chain + k - 1 && p > chain) {
2111        p->p--;
2112    } else {
2113        *top = *p->p;
2114        /* Nope, don't do this in ext3. Must leave the tree intact */
2115#if 0
2116        *p->p = 0;
2117#endif
2118    }
2119    /* Writer: end */
2120
2121    while(partial > p) {
2122        brelse(partial->bh);
2123        partial--;
2124    }
2125no_top:
2126    return partial;
2127}
2128
2129/*
2130 * Zero a number of block pointers in either an inode or an indirect block.
2131 * If we restart the transaction we must again get write access to the
2132 * indirect block for further modification.
2133 *
2134 * We release `count' blocks on disk, but (last - first) may be greater
2135 * than `count' because there can be holes in there.
2136 */
2137static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
2138        struct buffer_head *bh, ext3_fsblk_t block_to_free,
2139        unsigned long count, __le32 *first, __le32 *last)
2140{
2141    __le32 *p;
2142    if (try_to_extend_transaction(handle, inode)) {
2143        if (bh) {
2144            BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2145            if (ext3_journal_dirty_metadata(handle, bh))
2146                return;
2147        }
2148        ext3_mark_inode_dirty(handle, inode);
2149        truncate_restart_transaction(handle, inode);
2150        if (bh) {
2151            BUFFER_TRACE(bh, "retaking write access");
2152            if (ext3_journal_get_write_access(handle, bh))
2153                return;
2154        }
2155    }
2156
2157    /*
2158     * Any buffers which are on the journal will be in memory. We find
2159     * them on the hash table so journal_revoke() will run journal_forget()
2160     * on them. We've already detached each block from the file, so
2161     * bforget() in journal_forget() should be safe.
2162     *
2163     * AKPM: turn on bforget in journal_forget()!!!
2164     */
2165    for (p = first; p < last; p++) {
2166        u32 nr = le32_to_cpu(*p);
2167        if (nr) {
2168            struct buffer_head *bh;
2169
2170            *p = 0;
2171            bh = sb_find_get_block(inode->i_sb, nr);
2172            ext3_forget(handle, 0, inode, bh, nr);
2173        }
2174    }
2175
2176    ext3_free_blocks(handle, inode, block_to_free, count);
2177}
2178
2179/**
2180 * ext3_free_data - free a list of data blocks
2181 * @handle: handle for this transaction
2182 * @inode: inode we are dealing with
2183 * @this_bh: indirect buffer_head which contains *@first and *@last
2184 * @first: array of block numbers
2185 * @last: points immediately past the end of array
2186 *
2187 * We are freeing all blocks referred from that array (numbers are stored as
2188 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2189 *
2190 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2191 * blocks are contiguous then releasing them at one time will only affect one
2192 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2193 * actually use a lot of journal space.
2194 *
2195 * @this_bh will be %NULL if @first and @last point into the inode's direct
2196 * block pointers.
2197 */
2198static void ext3_free_data(handle_t *handle, struct inode *inode,
2199               struct buffer_head *this_bh,
2200               __le32 *first, __le32 *last)
2201{
2202    ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
2203    unsigned long count = 0; /* Number of blocks in the run */
2204    __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2205                           corresponding to
2206                           block_to_free */
2207    ext3_fsblk_t nr; /* Current block # */
2208    __le32 *p; /* Pointer into inode/ind
2209                           for current block */
2210    int err;
2211
2212    if (this_bh) { /* For indirect block */
2213        BUFFER_TRACE(this_bh, "get_write_access");
2214        err = ext3_journal_get_write_access(handle, this_bh);
2215        /* Important: if we can't update the indirect pointers
2216         * to the blocks, we can't free them. */
2217        if (err)
2218            return;
2219    }
2220
2221    for (p = first; p < last; p++) {
2222        nr = le32_to_cpu(*p);
2223        if (nr) {
2224            /* accumulate blocks to free if they're contiguous */
2225            if (count == 0) {
2226                block_to_free = nr;
2227                block_to_free_p = p;
2228                count = 1;
2229            } else if (nr == block_to_free + count) {
2230                count++;
2231            } else {
2232                ext3_clear_blocks(handle, inode, this_bh,
2233                          block_to_free,
2234                          count, block_to_free_p, p);
2235                block_to_free = nr;
2236                block_to_free_p = p;
2237                count = 1;
2238            }
2239        }
2240    }
2241
2242    if (count > 0)
2243        ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2244                  count, block_to_free_p, p);
2245
2246    if (this_bh) {
2247        BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2248
2249        /*
2250         * The buffer head should have an attached journal head at this
2251         * point. However, if the data is corrupted and an indirect
2252         * block pointed to itself, it would have been detached when
2253         * the block was cleared. Check for this instead of OOPSing.
2254         */
2255        if (bh2jh(this_bh))
2256            ext3_journal_dirty_metadata(handle, this_bh);
2257        else
2258            ext3_error(inode->i_sb, "ext3_free_data",
2259                   "circular indirect block detected, "
2260                   "inode=%lu, block=%llu",
2261                   inode->i_ino,
2262                   (unsigned long long)this_bh->b_blocknr);
2263    }
2264}
2265
2266/**
2267 * ext3_free_branches - free an array of branches
2268 * @handle: JBD handle for this transaction
2269 * @inode: inode we are dealing with
2270 * @parent_bh: the buffer_head which contains *@first and *@last
2271 * @first: array of block numbers
2272 * @last: pointer immediately past the end of array
2273 * @depth: depth of the branches to free
2274 *
2275 * We are freeing all blocks referred from these branches (numbers are
2276 * stored as little-endian 32-bit) and updating @inode->i_blocks
2277 * appropriately.
2278 */
2279static void ext3_free_branches(handle_t *handle, struct inode *inode,
2280                   struct buffer_head *parent_bh,
2281                   __le32 *first, __le32 *last, int depth)
2282{
2283    ext3_fsblk_t nr;
2284    __le32 *p;
2285
2286    if (is_handle_aborted(handle))
2287        return;
2288
2289    if (depth--) {
2290        struct buffer_head *bh;
2291        int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2292        p = last;
2293        while (--p >= first) {
2294            nr = le32_to_cpu(*p);
2295            if (!nr)
2296                continue; /* A hole */
2297
2298            /* Go read the buffer for the next level down */
2299            bh = sb_bread(inode->i_sb, nr);
2300
2301            /*
2302             * A read failure? Report error and clear slot
2303             * (should be rare).
2304             */
2305            if (!bh) {
2306                ext3_error(inode->i_sb, "ext3_free_branches",
2307                       "Read failure, inode=%lu, block="E3FSBLK,
2308                       inode->i_ino, nr);
2309                continue;
2310            }
2311
2312            /* This zaps the entire block. Bottom up. */
2313            BUFFER_TRACE(bh, "free child branches");
2314            ext3_free_branches(handle, inode, bh,
2315                       (__le32*)bh->b_data,
2316                       (__le32*)bh->b_data + addr_per_block,
2317                       depth);
2318
2319            /*
2320             * Everything below this this pointer has been
2321             * released. Now let this top-of-subtree go.
2322             *
2323             * We want the freeing of this indirect block to be
2324             * atomic in the journal with the updating of the
2325             * bitmap block which owns it. So make some room in
2326             * the journal.
2327             *
2328             * We zero the parent pointer *after* freeing its
2329             * pointee in the bitmaps, so if extend_transaction()
2330             * for some reason fails to put the bitmap changes and
2331             * the release into the same transaction, recovery
2332             * will merely complain about releasing a free block,
2333             * rather than leaking blocks.
2334             */
2335            if (is_handle_aborted(handle))
2336                return;
2337            if (try_to_extend_transaction(handle, inode)) {
2338                ext3_mark_inode_dirty(handle, inode);
2339                truncate_restart_transaction(handle, inode);
2340            }
2341
2342            /*
2343             * We've probably journalled the indirect block several
2344             * times during the truncate. But it's no longer
2345             * needed and we now drop it from the transaction via
2346             * journal_revoke().
2347             *
2348             * That's easy if it's exclusively part of this
2349             * transaction. But if it's part of the committing
2350             * transaction then journal_forget() will simply
2351             * brelse() it. That means that if the underlying
2352             * block is reallocated in ext3_get_block(),
2353             * unmap_underlying_metadata() will find this block
2354             * and will try to get rid of it. damn, damn. Thus
2355             * we don't allow a block to be reallocated until
2356             * a transaction freeing it has fully committed.
2357             *
2358             * We also have to make sure journal replay after a
2359             * crash does not overwrite non-journaled data blocks
2360             * with old metadata when the block got reallocated for
2361             * data. Thus we have to store a revoke record for a
2362             * block in the same transaction in which we free the
2363             * block.
2364             */
2365            ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2366
2367            ext3_free_blocks(handle, inode, nr, 1);
2368
2369            if (parent_bh) {
2370                /*
2371                 * The block which we have just freed is
2372                 * pointed to by an indirect block: journal it
2373                 */
2374                BUFFER_TRACE(parent_bh, "get_write_access");
2375                if (!ext3_journal_get_write_access(handle,
2376                                   parent_bh)){
2377                    *p = 0;
2378                    BUFFER_TRACE(parent_bh,
2379                    "call ext3_journal_dirty_metadata");
2380                    ext3_journal_dirty_metadata(handle,
2381                                    parent_bh);
2382                }
2383            }
2384        }
2385    } else {
2386        /* We have reached the bottom of the tree. */
2387        BUFFER_TRACE(parent_bh, "free data blocks");
2388        ext3_free_data(handle, inode, parent_bh, first, last);
2389    }
2390}
2391
2392int ext3_can_truncate(struct inode *inode)
2393{
2394    if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2395        return 0;
2396    if (S_ISREG(inode->i_mode))
2397        return 1;
2398    if (S_ISDIR(inode->i_mode))
2399        return 1;
2400    if (S_ISLNK(inode->i_mode))
2401        return !ext3_inode_is_fast_symlink(inode);
2402    return 0;
2403}
2404
2405/*
2406 * ext3_truncate()
2407 *
2408 * We block out ext3_get_block() block instantiations across the entire
2409 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2410 * simultaneously on behalf of the same inode.
2411 *
2412 * As we work through the truncate and commmit bits of it to the journal there
2413 * is one core, guiding principle: the file's tree must always be consistent on
2414 * disk. We must be able to restart the truncate after a crash.
2415 *
2416 * The file's tree may be transiently inconsistent in memory (although it
2417 * probably isn't), but whenever we close off and commit a journal transaction,
2418 * the contents of (the filesystem + the journal) must be consistent and
2419 * restartable. It's pretty simple, really: bottom up, right to left (although
2420 * left-to-right works OK too).
2421 *
2422 * Note that at recovery time, journal replay occurs *before* the restart of
2423 * truncate against the orphan inode list.
2424 *
2425 * The committed inode has the new, desired i_size (which is the same as
2426 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2427 * that this inode's truncate did not complete and it will again call
2428 * ext3_truncate() to have another go. So there will be instantiated blocks
2429 * to the right of the truncation point in a crashed ext3 filesystem. But
2430 * that's fine - as long as they are linked from the inode, the post-crash
2431 * ext3_truncate() run will find them and release them.
2432 */
2433void ext3_truncate(struct inode *inode)
2434{
2435    handle_t *handle;
2436    struct ext3_inode_info *ei = EXT3_I(inode);
2437    __le32 *i_data = ei->i_data;
2438    int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2439    struct address_space *mapping = inode->i_mapping;
2440    int offsets[4];
2441    Indirect chain[4];
2442    Indirect *partial;
2443    __le32 nr = 0;
2444    int n;
2445    long last_block;
2446    unsigned blocksize = inode->i_sb->s_blocksize;
2447    struct page *page;
2448
2449    if (!ext3_can_truncate(inode))
2450        goto out_notrans;
2451
2452    if (inode->i_size == 0 && ext3_should_writeback_data(inode))
2453        ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
2454
2455    /*
2456     * We have to lock the EOF page here, because lock_page() nests
2457     * outside journal_start().
2458     */
2459    if ((inode->i_size & (blocksize - 1)) == 0) {
2460        /* Block boundary? Nothing to do */
2461        page = NULL;
2462    } else {
2463        page = grab_cache_page(mapping,
2464                inode->i_size >> PAGE_CACHE_SHIFT);
2465        if (!page)
2466            goto out_notrans;
2467    }
2468
2469    handle = start_transaction(inode);
2470    if (IS_ERR(handle)) {
2471        if (page) {
2472            clear_highpage(page);
2473            flush_dcache_page(page);
2474            unlock_page(page);
2475            page_cache_release(page);
2476        }
2477        goto out_notrans;
2478    }
2479
2480    last_block = (inode->i_size + blocksize-1)
2481                    >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2482
2483    if (page)
2484        ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2485
2486    n = ext3_block_to_path(inode, last_block, offsets, NULL);
2487    if (n == 0)
2488        goto out_stop; /* error */
2489
2490    /*
2491     * OK. This truncate is going to happen. We add the inode to the
2492     * orphan list, so that if this truncate spans multiple transactions,
2493     * and we crash, we will resume the truncate when the filesystem
2494     * recovers. It also marks the inode dirty, to catch the new size.
2495     *
2496     * Implication: the file must always be in a sane, consistent
2497     * truncatable state while each transaction commits.
2498     */
2499    if (ext3_orphan_add(handle, inode))
2500        goto out_stop;
2501
2502    /*
2503     * The orphan list entry will now protect us from any crash which
2504     * occurs before the truncate completes, so it is now safe to propagate
2505     * the new, shorter inode size (held for now in i_size) into the
2506     * on-disk inode. We do this via i_disksize, which is the value which
2507     * ext3 *really* writes onto the disk inode.
2508     */
2509    ei->i_disksize = inode->i_size;
2510
2511    /*
2512     * From here we block out all ext3_get_block() callers who want to
2513     * modify the block allocation tree.
2514     */
2515    mutex_lock(&ei->truncate_mutex);
2516
2517    if (n == 1) { /* direct blocks */
2518        ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2519                   i_data + EXT3_NDIR_BLOCKS);
2520        goto do_indirects;
2521    }
2522
2523    partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2524    /* Kill the top of shared branch (not detached) */
2525    if (nr) {
2526        if (partial == chain) {
2527            /* Shared branch grows from the inode */
2528            ext3_free_branches(handle, inode, NULL,
2529                       &nr, &nr+1, (chain+n-1) - partial);
2530            *partial->p = 0;
2531            /*
2532             * We mark the inode dirty prior to restart,
2533             * and prior to stop. No need for it here.
2534             */
2535        } else {
2536            /* Shared branch grows from an indirect block */
2537            ext3_free_branches(handle, inode, partial->bh,
2538                    partial->p,
2539                    partial->p+1, (chain+n-1) - partial);
2540        }
2541    }
2542    /* Clear the ends of indirect blocks on the shared branch */
2543    while (partial > chain) {
2544        ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2545                   (__le32*)partial->bh->b_data+addr_per_block,
2546                   (chain+n-1) - partial);
2547        BUFFER_TRACE(partial->bh, "call brelse");
2548        brelse (partial->bh);
2549        partial--;
2550    }
2551do_indirects:
2552    /* Kill the remaining (whole) subtrees */
2553    switch (offsets[0]) {
2554    default:
2555        nr = i_data[EXT3_IND_BLOCK];
2556        if (nr) {
2557            ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2558            i_data[EXT3_IND_BLOCK] = 0;
2559        }
2560    case EXT3_IND_BLOCK:
2561        nr = i_data[EXT3_DIND_BLOCK];
2562        if (nr) {
2563            ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2564            i_data[EXT3_DIND_BLOCK] = 0;
2565        }
2566    case EXT3_DIND_BLOCK:
2567        nr = i_data[EXT3_TIND_BLOCK];
2568        if (nr) {
2569            ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2570            i_data[EXT3_TIND_BLOCK] = 0;
2571        }
2572    case EXT3_TIND_BLOCK:
2573        ;
2574    }
2575
2576    ext3_discard_reservation(inode);
2577
2578    mutex_unlock(&ei->truncate_mutex);
2579    inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2580    ext3_mark_inode_dirty(handle, inode);
2581
2582    /*
2583     * In a multi-transaction truncate, we only make the final transaction
2584     * synchronous
2585     */
2586    if (IS_SYNC(inode))
2587        handle->h_sync = 1;
2588out_stop:
2589    /*
2590     * If this was a simple ftruncate(), and the file will remain alive
2591     * then we need to clear up the orphan record which we created above.
2592     * However, if this was a real unlink then we were called by
2593     * ext3_evict_inode(), and we allow that function to clean up the
2594     * orphan info for us.
2595     */
2596    if (inode->i_nlink)
2597        ext3_orphan_del(handle, inode);
2598
2599    ext3_journal_stop(handle);
2600    return;
2601out_notrans:
2602    /*
2603     * Delete the inode from orphan list so that it doesn't stay there
2604     * forever and trigger assertion on umount.
2605     */
2606    if (inode->i_nlink)
2607        ext3_orphan_del(NULL, inode);
2608}
2609
2610static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
2611        unsigned long ino, struct ext3_iloc *iloc)
2612{
2613    unsigned long block_group;
2614    unsigned long offset;
2615    ext3_fsblk_t block;
2616    struct ext3_group_desc *gdp;
2617
2618    if (!ext3_valid_inum(sb, ino)) {
2619        /*
2620         * This error is already checked for in namei.c unless we are
2621         * looking at an NFS filehandle, in which case no error
2622         * report is needed
2623         */
2624        return 0;
2625    }
2626
2627    block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2628    gdp = ext3_get_group_desc(sb, block_group, NULL);
2629    if (!gdp)
2630        return 0;
2631    /*
2632     * Figure out the offset within the block group inode table
2633     */
2634    offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2635        EXT3_INODE_SIZE(sb);
2636    block = le32_to_cpu(gdp->bg_inode_table) +
2637        (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2638
2639    iloc->block_group = block_group;
2640    iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2641    return block;
2642}
2643
2644/*
2645 * ext3_get_inode_loc returns with an extra refcount against the inode's
2646 * underlying buffer_head on success. If 'in_mem' is true, we have all
2647 * data in memory that is needed to recreate the on-disk version of this
2648 * inode.
2649 */
2650static int __ext3_get_inode_loc(struct inode *inode,
2651                struct ext3_iloc *iloc, int in_mem)
2652{
2653    ext3_fsblk_t block;
2654    struct buffer_head *bh;
2655
2656    block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2657    if (!block)
2658        return -EIO;
2659
2660    bh = sb_getblk(inode->i_sb, block);
2661    if (!bh) {
2662        ext3_error (inode->i_sb, "ext3_get_inode_loc",
2663                "unable to read inode block - "
2664                "inode=%lu, block="E3FSBLK,
2665                 inode->i_ino, block);
2666        return -EIO;
2667    }
2668    if (!buffer_uptodate(bh)) {
2669        lock_buffer(bh);
2670
2671        /*
2672         * If the buffer has the write error flag, we have failed
2673         * to write out another inode in the same block. In this
2674         * case, we don't have to read the block because we may
2675         * read the old inode data successfully.
2676         */
2677        if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2678            set_buffer_uptodate(bh);
2679
2680        if (buffer_uptodate(bh)) {
2681            /* someone brought it uptodate while we waited */
2682            unlock_buffer(bh);
2683            goto has_buffer;
2684        }
2685
2686        /*
2687         * If we have all information of the inode in memory and this
2688         * is the only valid inode in the block, we need not read the
2689         * block.
2690         */
2691        if (in_mem) {
2692            struct buffer_head *bitmap_bh;
2693            struct ext3_group_desc *desc;
2694            int inodes_per_buffer;
2695            int inode_offset, i;
2696            int block_group;
2697            int start;
2698
2699            block_group = (inode->i_ino - 1) /
2700                    EXT3_INODES_PER_GROUP(inode->i_sb);
2701            inodes_per_buffer = bh->b_size /
2702                EXT3_INODE_SIZE(inode->i_sb);
2703            inode_offset = ((inode->i_ino - 1) %
2704                    EXT3_INODES_PER_GROUP(inode->i_sb));
2705            start = inode_offset & ~(inodes_per_buffer - 1);
2706
2707            /* Is the inode bitmap in cache? */
2708            desc = ext3_get_group_desc(inode->i_sb,
2709                        block_group, NULL);
2710            if (!desc)
2711                goto make_io;
2712
2713            bitmap_bh = sb_getblk(inode->i_sb,
2714                    le32_to_cpu(desc->bg_inode_bitmap));
2715            if (!bitmap_bh)
2716                goto make_io;
2717
2718            /*
2719             * If the inode bitmap isn't in cache then the
2720             * optimisation may end up performing two reads instead
2721             * of one, so skip it.
2722             */
2723            if (!buffer_uptodate(bitmap_bh)) {
2724                brelse(bitmap_bh);
2725                goto make_io;
2726            }
2727            for (i = start; i < start + inodes_per_buffer; i++) {
2728                if (i == inode_offset)
2729                    continue;
2730                if (ext3_test_bit(i, bitmap_bh->b_data))
2731                    break;
2732            }
2733            brelse(bitmap_bh);
2734            if (i == start + inodes_per_buffer) {
2735                /* all other inodes are free, so skip I/O */
2736                memset(bh->b_data, 0, bh->b_size);
2737                set_buffer_uptodate(bh);
2738                unlock_buffer(bh);
2739                goto has_buffer;
2740            }
2741        }
2742
2743make_io:
2744        /*
2745         * There are other valid inodes in the buffer, this inode
2746         * has in-inode xattrs, or we don't have this inode in memory.
2747         * Read the block from disk.
2748         */
2749        get_bh(bh);
2750        bh->b_end_io = end_buffer_read_sync;
2751        submit_bh(READ_META, bh);
2752        wait_on_buffer(bh);
2753        if (!buffer_uptodate(bh)) {
2754            ext3_error(inode->i_sb, "ext3_get_inode_loc",
2755                    "unable to read inode block - "
2756                    "inode=%lu, block="E3FSBLK,
2757                    inode->i_ino, block);
2758            brelse(bh);
2759            return -EIO;
2760        }
2761    }
2762has_buffer:
2763    iloc->bh = bh;
2764    return 0;
2765}
2766
2767int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2768{
2769    /* We have all inode data except xattrs in memory here. */
2770    return __ext3_get_inode_loc(inode, iloc,
2771        !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
2772}
2773
2774void ext3_set_inode_flags(struct inode *inode)
2775{
2776    unsigned int flags = EXT3_I(inode)->i_flags;
2777
2778    inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2779    if (flags & EXT3_SYNC_FL)
2780        inode->i_flags |= S_SYNC;
2781    if (flags & EXT3_APPEND_FL)
2782        inode->i_flags |= S_APPEND;
2783    if (flags & EXT3_IMMUTABLE_FL)
2784        inode->i_flags |= S_IMMUTABLE;
2785    if (flags & EXT3_NOATIME_FL)
2786        inode->i_flags |= S_NOATIME;
2787    if (flags & EXT3_DIRSYNC_FL)
2788        inode->i_flags |= S_DIRSYNC;
2789}
2790
2791/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2792void ext3_get_inode_flags(struct ext3_inode_info *ei)
2793{
2794    unsigned int flags = ei->vfs_inode.i_flags;
2795
2796    ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2797            EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2798    if (flags & S_SYNC)
2799        ei->i_flags |= EXT3_SYNC_FL;
2800    if (flags & S_APPEND)
2801        ei->i_flags |= EXT3_APPEND_FL;
2802    if (flags & S_IMMUTABLE)
2803        ei->i_flags |= EXT3_IMMUTABLE_FL;
2804    if (flags & S_NOATIME)
2805        ei->i_flags |= EXT3_NOATIME_FL;
2806    if (flags & S_DIRSYNC)
2807        ei->i_flags |= EXT3_DIRSYNC_FL;
2808}
2809
2810struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
2811{
2812    struct ext3_iloc iloc;
2813    struct ext3_inode *raw_inode;
2814    struct ext3_inode_info *ei;
2815    struct buffer_head *bh;
2816    struct inode *inode;
2817    journal_t *journal = EXT3_SB(sb)->s_journal;
2818    transaction_t *transaction;
2819    long ret;
2820    int block;
2821
2822    inode = iget_locked(sb, ino);
2823    if (!inode)
2824        return ERR_PTR(-ENOMEM);
2825    if (!(inode->i_state & I_NEW))
2826        return inode;
2827
2828    ei = EXT3_I(inode);
2829    ei->i_block_alloc_info = NULL;
2830
2831    ret = __ext3_get_inode_loc(inode, &iloc, 0);
2832    if (ret < 0)
2833        goto bad_inode;
2834    bh = iloc.bh;
2835    raw_inode = ext3_raw_inode(&iloc);
2836    inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2837    inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2838    inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2839    if(!(test_opt (inode->i_sb, NO_UID32))) {
2840        inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2841        inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2842    }
2843    inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2844    inode->i_size = le32_to_cpu(raw_inode->i_size);
2845    inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2846    inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2847    inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
2848    inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2849
2850    ei->i_state_flags = 0;
2851    ei->i_dir_start_lookup = 0;
2852    ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2853    /* We now have enough fields to check if the inode was active or not.
2854     * This is needed because nfsd might try to access dead inodes
2855     * the test is that same one that e2fsck uses
2856     * NeilBrown 1999oct15
2857     */
2858    if (inode->i_nlink == 0) {
2859        if (inode->i_mode == 0 ||
2860            !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2861            /* this inode is deleted */
2862            brelse (bh);
2863            ret = -ESTALE;
2864            goto bad_inode;
2865        }
2866        /* The only unlinked inodes we let through here have
2867         * valid i_mode and are being read by the orphan
2868         * recovery code: that's fine, we're about to complete
2869         * the process of deleting those. */
2870    }
2871    inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2872    ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2873#ifdef EXT3_FRAGMENTS
2874    ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2875    ei->i_frag_no = raw_inode->i_frag;
2876    ei->i_frag_size = raw_inode->i_fsize;
2877#endif
2878    ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2879    if (!S_ISREG(inode->i_mode)) {
2880        ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2881    } else {
2882        inode->i_size |=
2883            ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2884    }
2885    ei->i_disksize = inode->i_size;
2886    inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2887    ei->i_block_group = iloc.block_group;
2888    /*
2889     * NOTE! The in-memory inode i_data array is in little-endian order
2890     * even on big-endian machines: we do NOT byteswap the block numbers!
2891     */
2892    for (block = 0; block < EXT3_N_BLOCKS; block++)
2893        ei->i_data[block] = raw_inode->i_block[block];
2894    INIT_LIST_HEAD(&ei->i_orphan);
2895
2896    /*
2897     * Set transaction id's of transactions that have to be committed
2898     * to finish f[data]sync. We set them to currently running transaction
2899     * as we cannot be sure that the inode or some of its metadata isn't
2900     * part of the transaction - the inode could have been reclaimed and
2901     * now it is reread from disk.
2902     */
2903    if (journal) {
2904        tid_t tid;
2905
2906        spin_lock(&journal->j_state_lock);
2907        if (journal->j_running_transaction)
2908            transaction = journal->j_running_transaction;
2909        else
2910            transaction = journal->j_committing_transaction;
2911        if (transaction)
2912            tid = transaction->t_tid;
2913        else
2914            tid = journal->j_commit_sequence;
2915        spin_unlock(&journal->j_state_lock);
2916        atomic_set(&ei->i_sync_tid, tid);
2917        atomic_set(&ei->i_datasync_tid, tid);
2918    }
2919
2920    if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2921        EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2922        /*
2923         * When mke2fs creates big inodes it does not zero out
2924         * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2925         * so ignore those first few inodes.
2926         */
2927        ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2928        if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2929            EXT3_INODE_SIZE(inode->i_sb)) {
2930            brelse (bh);
2931            ret = -EIO;
2932            goto bad_inode;
2933        }
2934        if (ei->i_extra_isize == 0) {
2935            /* The extra space is currently unused. Use it. */
2936            ei->i_extra_isize = sizeof(struct ext3_inode) -
2937                        EXT3_GOOD_OLD_INODE_SIZE;
2938        } else {
2939            __le32 *magic = (void *)raw_inode +
2940                    EXT3_GOOD_OLD_INODE_SIZE +
2941                    ei->i_extra_isize;
2942            if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2943                 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
2944        }
2945    } else
2946        ei->i_extra_isize = 0;
2947
2948    if (S_ISREG(inode->i_mode)) {
2949        inode->i_op = &ext3_file_inode_operations;
2950        inode->i_fop = &ext3_file_operations;
2951        ext3_set_aops(inode);
2952    } else if (S_ISDIR(inode->i_mode)) {
2953        inode->i_op = &ext3_dir_inode_operations;
2954        inode->i_fop = &ext3_dir_operations;
2955    } else if (S_ISLNK(inode->i_mode)) {
2956        if (ext3_inode_is_fast_symlink(inode)) {
2957            inode->i_op = &ext3_fast_symlink_inode_operations;
2958            nd_terminate_link(ei->i_data, inode->i_size,
2959                sizeof(ei->i_data) - 1);
2960        } else {
2961            inode->i_op = &ext3_symlink_inode_operations;
2962            ext3_set_aops(inode);
2963        }
2964    } else {
2965        inode->i_op = &ext3_special_inode_operations;
2966        if (raw_inode->i_block[0])
2967            init_special_inode(inode, inode->i_mode,
2968               old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2969        else
2970            init_special_inode(inode, inode->i_mode,
2971               new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2972    }
2973    brelse (iloc.bh);
2974    ext3_set_inode_flags(inode);
2975    unlock_new_inode(inode);
2976    return inode;
2977
2978bad_inode:
2979    iget_failed(inode);
2980    return ERR_PTR(ret);
2981}
2982
2983/*
2984 * Post the struct inode info into an on-disk inode location in the
2985 * buffer-cache. This gobbles the caller's reference to the
2986 * buffer_head in the inode location struct.
2987 *
2988 * The caller must have write access to iloc->bh.
2989 */
2990static int ext3_do_update_inode(handle_t *handle,
2991                struct inode *inode,
2992                struct ext3_iloc *iloc)
2993{
2994    struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2995    struct ext3_inode_info *ei = EXT3_I(inode);
2996    struct buffer_head *bh = iloc->bh;
2997    int err = 0, rc, block;
2998
2999again:
3000    /* we can't allow multiple procs in here at once, its a bit racey */
3001    lock_buffer(bh);
3002
3003    /* For fields not not tracking in the in-memory inode,
3004     * initialise them to zero for new inodes. */
3005    if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
3006        memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3007
3008    ext3_get_inode_flags(ei);
3009    raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3010    if(!(test_opt(inode->i_sb, NO_UID32))) {
3011        raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3012        raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3013/*
3014 * Fix up interoperability with old kernels. Otherwise, old inodes get
3015 * re-used with the upper 16 bits of the uid/gid intact
3016 */
3017        if(!ei->i_dtime) {
3018            raw_inode->i_uid_high =
3019                cpu_to_le16(high_16_bits(inode->i_uid));
3020            raw_inode->i_gid_high =
3021                cpu_to_le16(high_16_bits(inode->i_gid));
3022        } else {
3023            raw_inode->i_uid_high = 0;
3024            raw_inode->i_gid_high = 0;
3025        }
3026    } else {
3027        raw_inode->i_uid_low =
3028            cpu_to_le16(fs_high2lowuid(inode->i_uid));
3029        raw_inode->i_gid_low =
3030            cpu_to_le16(fs_high2lowgid(inode->i_gid));
3031        raw_inode->i_uid_high = 0;
3032        raw_inode->i_gid_high = 0;
3033    }
3034    raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3035    raw_inode->i_size = cpu_to_le32(ei->i_disksize);
3036    raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3037    raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3038    raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3039    raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3040    raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3041    raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3042#ifdef EXT3_FRAGMENTS
3043    raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3044    raw_inode->i_frag = ei->i_frag_no;
3045    raw_inode->i_fsize = ei->i_frag_size;
3046#endif
3047    raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3048    if (!S_ISREG(inode->i_mode)) {
3049        raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3050    } else {
3051        raw_inode->i_size_high =
3052            cpu_to_le32(ei->i_disksize >> 32);
3053        if (ei->i_disksize > 0x7fffffffULL) {
3054            struct super_block *sb = inode->i_sb;
3055            if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3056                    EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3057                EXT3_SB(sb)->s_es->s_rev_level ==
3058                    cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3059                   /* If this is the first large file
3060                * created, add a flag to the superblock.
3061                */
3062                unlock_buffer(bh);
3063                err = ext3_journal_get_write_access(handle,
3064                        EXT3_SB(sb)->s_sbh);
3065                if (err)
3066                    goto out_brelse;
3067
3068                ext3_update_dynamic_rev(sb);
3069                EXT3_SET_RO_COMPAT_FEATURE(sb,
3070                    EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
3071                handle->h_sync = 1;
3072                err = ext3_journal_dirty_metadata(handle,
3073                        EXT3_SB(sb)->s_sbh);
3074                /* get our lock and start over */
3075                goto again;
3076            }
3077        }
3078    }
3079    raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3080    if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3081        if (old_valid_dev(inode->i_rdev)) {
3082            raw_inode->i_block[0] =
3083                cpu_to_le32(old_encode_dev(inode->i_rdev));
3084            raw_inode->i_block[1] = 0;
3085        } else {
3086            raw_inode->i_block[0] = 0;
3087            raw_inode->i_block[1] =
3088                cpu_to_le32(new_encode_dev(inode->i_rdev));
3089            raw_inode->i_block[2] = 0;
3090        }
3091    } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3092        raw_inode->i_block[block] = ei->i_data[block];
3093
3094    if (ei->i_extra_isize)
3095        raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3096
3097    BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
3098    unlock_buffer(bh);
3099    rc = ext3_journal_dirty_metadata(handle, bh);
3100    if (!err)
3101        err = rc;
3102    ext3_clear_inode_state(inode, EXT3_STATE_NEW);
3103
3104    atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
3105out_brelse:
3106    brelse (bh);
3107    ext3_std_error(inode->i_sb, err);
3108    return err;
3109}
3110
3111/*
3112 * ext3_write_inode()
3113 *
3114 * We are called from a few places:
3115 *
3116 * - Within generic_file_write() for O_SYNC files.
3117 * Here, there will be no transaction running. We wait for any running
3118 * trasnaction to commit.
3119 *
3120 * - Within sys_sync(), kupdate and such.
3121 * We wait on commit, if tol to.
3122 *
3123 * - Within prune_icache() (PF_MEMALLOC == true)
3124 * Here we simply return. We can't afford to block kswapd on the
3125 * journal commit.
3126 *
3127 * In all cases it is actually safe for us to return without doing anything,
3128 * because the inode has been copied into a raw inode buffer in
3129 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3130 * knfsd.
3131 *
3132 * Note that we are absolutely dependent upon all inode dirtiers doing the
3133 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3134 * which we are interested.
3135 *
3136 * It would be a bug for them to not do this. The code:
3137 *
3138 * mark_inode_dirty(inode)
3139 * stuff();
3140 * inode->i_size = expr;
3141 *
3142 * is in error because a kswapd-driven write_inode() could occur while
3143 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3144 * will no longer be on the superblock's dirty inode list.
3145 */
3146int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
3147{
3148    if (current->flags & PF_MEMALLOC)
3149        return 0;
3150
3151    if (ext3_journal_current_handle()) {
3152        jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3153        dump_stack();
3154        return -EIO;
3155    }
3156
3157    if (wbc->sync_mode != WB_SYNC_ALL)
3158        return 0;
3159
3160    return ext3_force_commit(inode->i_sb);
3161}
3162
3163/*
3164 * ext3_setattr()
3165 *
3166 * Called from notify_change.
3167 *
3168 * We want to trap VFS attempts to truncate the file as soon as
3169 * possible. In particular, we want to make sure that when the VFS
3170 * shrinks i_size, we put the inode on the orphan list and modify
3171 * i_disksize immediately, so that during the subsequent flushing of
3172 * dirty pages and freeing of disk blocks, we can guarantee that any
3173 * commit will leave the blocks being flushed in an unused state on
3174 * disk. (On recovery, the inode will get truncated and the blocks will
3175 * be freed, so we have a strong guarantee that no future commit will
3176 * leave these blocks visible to the user.)
3177 *
3178 * Called with inode->sem down.
3179 */
3180int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3181{
3182    struct inode *inode = dentry->d_inode;
3183    int error, rc = 0;
3184    const unsigned int ia_valid = attr->ia_valid;
3185
3186    error = inode_change_ok(inode, attr);
3187    if (error)
3188        return error;
3189
3190    if (is_quota_modification(inode, attr))
3191        dquot_initialize(inode);
3192    if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3193        (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3194        handle_t *handle;
3195
3196        /* (user+group)*(old+new) structure, inode write (sb,
3197         * inode block, ? - but truncate inode update has it) */
3198        handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3199                    EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
3200        if (IS_ERR(handle)) {
3201            error = PTR_ERR(handle);
3202            goto err_out;
3203        }
3204        error = dquot_transfer(inode, attr);
3205        if (error) {
3206            ext3_journal_stop(handle);
3207            return error;
3208        }
3209        /* Update corresponding info in inode so that everything is in
3210         * one transaction */
3211        if (attr->ia_valid & ATTR_UID)
3212            inode->i_uid = attr->ia_uid;
3213        if (attr->ia_valid & ATTR_GID)
3214            inode->i_gid = attr->ia_gid;
3215        error = ext3_mark_inode_dirty(handle, inode);
3216        ext3_journal_stop(handle);
3217    }
3218
3219    if (S_ISREG(inode->i_mode) &&
3220        attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3221        handle_t *handle;
3222
3223        handle = ext3_journal_start(inode, 3);
3224        if (IS_ERR(handle)) {
3225            error = PTR_ERR(handle);
3226            goto err_out;
3227        }
3228
3229        error = ext3_orphan_add(handle, inode);
3230        EXT3_I(inode)->i_disksize = attr->ia_size;
3231        rc = ext3_mark_inode_dirty(handle, inode);
3232        if (!error)
3233            error = rc;
3234        ext3_journal_stop(handle);
3235    }
3236
3237    if ((attr->ia_valid & ATTR_SIZE) &&
3238        attr->ia_size != i_size_read(inode)) {
3239        rc = vmtruncate(inode, attr->ia_size);
3240        if (rc)
3241            goto err_out;
3242    }
3243
3244    setattr_copy(inode, attr);
3245    mark_inode_dirty(inode);
3246
3247    if (ia_valid & ATTR_MODE)
3248        rc = ext3_acl_chmod(inode);
3249
3250err_out:
3251    ext3_std_error(inode->i_sb, error);
3252    if (!error)
3253        error = rc;
3254    return error;
3255}
3256
3257
3258/*
3259 * How many blocks doth make a writepage()?
3260 *
3261 * With N blocks per page, it may be:
3262 * N data blocks
3263 * 2 indirect block
3264 * 2 dindirect
3265 * 1 tindirect
3266 * N+5 bitmap blocks (from the above)
3267 * N+5 group descriptor summary blocks
3268 * 1 inode block
3269 * 1 superblock.
3270 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3271 *
3272 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3273 *
3274 * With ordered or writeback data it's the same, less the N data blocks.
3275 *
3276 * If the inode's direct blocks can hold an integral number of pages then a
3277 * page cannot straddle two indirect blocks, and we can only touch one indirect
3278 * and dindirect block, and the "5" above becomes "3".
3279 *
3280 * This still overestimates under most circumstances. If we were to pass the
3281 * start and end offsets in here as well we could do block_to_path() on each
3282 * block and work out the exact number of indirects which are touched. Pah.
3283 */
3284
3285static int ext3_writepage_trans_blocks(struct inode *inode)
3286{
3287    int bpp = ext3_journal_blocks_per_page(inode);
3288    int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3289    int ret;
3290
3291    if (ext3_should_journal_data(inode))
3292        ret = 3 * (bpp + indirects) + 2;
3293    else
3294        ret = 2 * (bpp + indirects) + indirects + 2;
3295
3296#ifdef CONFIG_QUOTA
3297    /* We know that structure was already allocated during dquot_initialize so
3298     * we will be updating only the data blocks + inodes */
3299    ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
3300#endif
3301
3302    return ret;
3303}
3304
3305/*
3306 * The caller must have previously called ext3_reserve_inode_write().
3307 * Give this, we know that the caller already has write access to iloc->bh.
3308 */
3309int ext3_mark_iloc_dirty(handle_t *handle,
3310        struct inode *inode, struct ext3_iloc *iloc)
3311{
3312    int err = 0;
3313
3314    /* the do_update_inode consumes one bh->b_count */
3315    get_bh(iloc->bh);
3316
3317    /* ext3_do_update_inode() does journal_dirty_metadata */
3318    err = ext3_do_update_inode(handle, inode, iloc);
3319    put_bh(iloc->bh);
3320    return err;
3321}
3322
3323/*
3324 * On success, We end up with an outstanding reference count against
3325 * iloc->bh. This _must_ be cleaned up later.
3326 */
3327
3328int
3329ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
3330             struct ext3_iloc *iloc)
3331{
3332    int err = 0;
3333    if (handle) {
3334        err = ext3_get_inode_loc(inode, iloc);
3335        if (!err) {
3336            BUFFER_TRACE(iloc->bh, "get_write_access");
3337            err = ext3_journal_get_write_access(handle, iloc->bh);
3338            if (err) {
3339                brelse(iloc->bh);
3340                iloc->bh = NULL;
3341            }
3342        }
3343    }
3344    ext3_std_error(inode->i_sb, err);
3345    return err;
3346}
3347
3348/*
3349 * What we do here is to mark the in-core inode as clean with respect to inode
3350 * dirtiness (it may still be data-dirty).
3351 * This means that the in-core inode may be reaped by prune_icache
3352 * without having to perform any I/O. This is a very good thing,
3353 * because *any* task may call prune_icache - even ones which
3354 * have a transaction open against a different journal.
3355 *
3356 * Is this cheating? Not really. Sure, we haven't written the
3357 * inode out, but prune_icache isn't a user-visible syncing function.
3358 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3359 * we start and wait on commits.
3360 *
3361 * Is this efficient/effective? Well, we're being nice to the system
3362 * by cleaning up our inodes proactively so they can be reaped
3363 * without I/O. But we are potentially leaving up to five seconds'
3364 * worth of inodes floating about which prune_icache wants us to
3365 * write out. One way to fix that would be to get prune_icache()
3366 * to do a write_super() to free up some memory. It has the desired
3367 * effect.
3368 */
3369int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3370{
3371    struct ext3_iloc iloc;
3372    int err;
3373
3374    might_sleep();
3375    err = ext3_reserve_inode_write(handle, inode, &iloc);
3376    if (!err)
3377        err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3378    return err;
3379}
3380
3381/*
3382 * ext3_dirty_inode() is called from __mark_inode_dirty()
3383 *
3384 * We're really interested in the case where a file is being extended.
3385 * i_size has been changed by generic_commit_write() and we thus need
3386 * to include the updated inode in the current transaction.
3387 *
3388 * Also, dquot_alloc_space() will always dirty the inode when blocks
3389 * are allocated to the file.
3390 *
3391 * If the inode is marked synchronous, we don't honour that here - doing
3392 * so would cause a commit on atime updates, which we don't bother doing.
3393 * We handle synchronous inodes at the highest possible level.
3394 */
3395void ext3_dirty_inode(struct inode *inode, int flags)
3396{
3397    handle_t *current_handle = ext3_journal_current_handle();
3398    handle_t *handle;
3399
3400    handle = ext3_journal_start(inode, 2);
3401    if (IS_ERR(handle))
3402        goto out;
3403    if (current_handle &&
3404        current_handle->h_transaction != handle->h_transaction) {
3405        /* This task has a transaction open against a different fs */
3406        printk(KERN_EMERG "%s: transactions do not match!\n",
3407               __func__);
3408    } else {
3409        jbd_debug(5, "marking dirty. outer handle=%p\n",
3410                current_handle);
3411        ext3_mark_inode_dirty(handle, inode);
3412    }
3413    ext3_journal_stop(handle);
3414out:
3415    return;
3416}
3417
3418#if 0
3419/*
3420 * Bind an inode's backing buffer_head into this transaction, to prevent
3421 * it from being flushed to disk early. Unlike
3422 * ext3_reserve_inode_write, this leaves behind no bh reference and
3423 * returns no iloc structure, so the caller needs to repeat the iloc
3424 * lookup to mark the inode dirty later.
3425 */
3426static int ext3_pin_inode(handle_t *handle, struct inode *inode)
3427{
3428    struct ext3_iloc iloc;
3429
3430    int err = 0;
3431    if (handle) {
3432        err = ext3_get_inode_loc(inode, &iloc);
3433        if (!err) {
3434            BUFFER_TRACE(iloc.bh, "get_write_access");
3435            err = journal_get_write_access(handle, iloc.bh);
3436            if (!err)
3437                err = ext3_journal_dirty_metadata(handle,
3438                                  iloc.bh);
3439            brelse(iloc.bh);
3440        }
3441    }
3442    ext3_std_error(inode->i_sb, err);
3443    return err;
3444}
3445#endif
3446
3447int ext3_change_inode_journal_flag(struct inode *inode, int val)
3448{
3449    journal_t *journal;
3450    handle_t *handle;
3451    int err;
3452
3453    /*
3454     * We have to be very careful here: changing a data block's
3455     * journaling status dynamically is dangerous. If we write a
3456     * data block to the journal, change the status and then delete
3457     * that block, we risk forgetting to revoke the old log record
3458     * from the journal and so a subsequent replay can corrupt data.
3459     * So, first we make sure that the journal is empty and that
3460     * nobody is changing anything.
3461     */
3462
3463    journal = EXT3_JOURNAL(inode);
3464    if (is_journal_aborted(journal))
3465        return -EROFS;
3466
3467    journal_lock_updates(journal);
3468    journal_flush(journal);
3469
3470    /*
3471     * OK, there are no updates running now, and all cached data is
3472     * synced to disk. We are now in a completely consistent state
3473     * which doesn't have anything in the journal, and we know that
3474     * no filesystem updates are running, so it is safe to modify
3475     * the inode's in-core data-journaling state flag now.
3476     */
3477
3478    if (val)
3479        EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3480    else
3481        EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3482    ext3_set_aops(inode);
3483
3484    journal_unlock_updates(journal);
3485
3486    /* Finally we can mark the inode as dirty. */
3487
3488    handle = ext3_journal_start(inode, 1);
3489    if (IS_ERR(handle))
3490        return PTR_ERR(handle);
3491
3492    err = ext3_mark_inode_dirty(handle, inode);
3493    handle->h_sync = 1;
3494    ext3_journal_stop(handle);
3495    ext3_std_error(inode->i_sb, err);
3496
3497    return err;
3498}
3499

Archive Download this file



interactive