~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/gfs2/dir.c

Version: ~ [ linux-5.9-rc6 ] ~ [ linux-5.8.10 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.66 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.146 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.198 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.236 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.236 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
  4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
  5  */
  6 
  7 /*
  8  * Implements Extendible Hashing as described in:
  9  *   "Extendible Hashing" by Fagin, et al in
 10  *     __ACM Trans. on Database Systems__, Sept 1979.
 11  *
 12  *
 13  * Here's the layout of dirents which is essentially the same as that of ext2
 14  * within a single block. The field de_name_len is the number of bytes
 15  * actually required for the name (no null terminator). The field de_rec_len
 16  * is the number of bytes allocated to the dirent. The offset of the next
 17  * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
 18  * deleted, the preceding dirent inherits its allocated space, ie
 19  * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
 20  * by adding de_rec_len to the current dirent, this essentially causes the
 21  * deleted dirent to get jumped over when iterating through all the dirents.
 22  *
 23  * When deleting the first dirent in a block, there is no previous dirent so
 24  * the field de_ino is set to zero to designate it as deleted. When allocating
 25  * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
 26  * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
 27  * dirent is allocated. Otherwise it must go through all the 'used' dirents
 28  * searching for one in which the amount of total space minus the amount of
 29  * used space will provide enough space for the new dirent.
 30  *
 31  * There are two types of blocks in which dirents reside. In a stuffed dinode,
 32  * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
 33  * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
 34  * beginning of the leaf block. The dirents reside in leaves when
 35  *
 36  * dip->i_diskflags & GFS2_DIF_EXHASH is true
 37  *
 38  * Otherwise, the dirents are "linear", within a single stuffed dinode block.
 39  *
 40  * When the dirents are in leaves, the actual contents of the directory file are
 41  * used as an array of 64-bit block pointers pointing to the leaf blocks. The
 42  * dirents are NOT in the directory file itself. There can be more than one
 43  * block pointer in the array that points to the same leaf. In fact, when a
 44  * directory is first converted from linear to exhash, all of the pointers
 45  * point to the same leaf.
 46  *
 47  * When a leaf is completely full, the size of the hash table can be
 48  * doubled unless it is already at the maximum size which is hard coded into
 49  * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
 50  * but never before the maximum hash table size has been reached.
 51  */
 52 
 53 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 54 
 55 #include <linux/slab.h>
 56 #include <linux/spinlock.h>
 57 #include <linux/buffer_head.h>
 58 #include <linux/sort.h>
 59 #include <linux/gfs2_ondisk.h>
 60 #include <linux/crc32.h>
 61 #include <linux/vmalloc.h>
 62 #include <linux/bio.h>
 63 
 64 #include "gfs2.h"
 65 #include "incore.h"
 66 #include "dir.h"
 67 #include "glock.h"
 68 #include "inode.h"
 69 #include "meta_io.h"
 70 #include "quota.h"
 71 #include "rgrp.h"
 72 #include "trans.h"
 73 #include "bmap.h"
 74 #include "util.h"
 75 
 76 #define IS_LEAF     1 /* Hashed (leaf) directory */
 77 #define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
 78 
 79 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
 80 
 81 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
 82 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
 83 #define GFS2_HASH_INDEX_MASK 0xffffc000
 84 #define GFS2_USE_HASH_FLAG 0x2000
 85 
 86 struct qstr gfs2_qdot __read_mostly;
 87 struct qstr gfs2_qdotdot __read_mostly;
 88 
 89 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
 90                             const struct qstr *name, void *opaque);
 91 
 92 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
 93                             struct buffer_head **bhp)
 94 {
 95         struct buffer_head *bh;
 96 
 97         bh = gfs2_meta_new(ip->i_gl, block);
 98         gfs2_trans_add_meta(ip->i_gl, bh);
 99         gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
100         gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
101         *bhp = bh;
102         return 0;
103 }
104 
105 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
106                                         struct buffer_head **bhp)
107 {
108         struct buffer_head *bh;
109         int error;
110 
111         error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
112         if (error)
113                 return error;
114         if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
115                 brelse(bh);
116                 return -EIO;
117         }
118         *bhp = bh;
119         return 0;
120 }
121 
122 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
123                                   unsigned int offset, unsigned int size)
124 {
125         struct buffer_head *dibh;
126         int error;
127 
128         error = gfs2_meta_inode_buffer(ip, &dibh);
129         if (error)
130                 return error;
131 
132         gfs2_trans_add_meta(ip->i_gl, dibh);
133         memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
134         if (ip->i_inode.i_size < offset + size)
135                 i_size_write(&ip->i_inode, offset + size);
136         ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
137         gfs2_dinode_out(ip, dibh->b_data);
138 
139         brelse(dibh);
140 
141         return size;
142 }
143 
144 
145 
146 /**
147  * gfs2_dir_write_data - Write directory information to the inode
148  * @ip: The GFS2 inode
149  * @buf: The buffer containing information to be written
150  * @offset: The file offset to start writing at
151  * @size: The amount of data to write
152  *
153  * Returns: The number of bytes correctly written or error code
154  */
155 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
156                                u64 offset, unsigned int size)
157 {
158         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
159         struct buffer_head *dibh;
160         u64 lblock, dblock;
161         u32 extlen = 0;
162         unsigned int o;
163         int copied = 0;
164         int error = 0;
165         int new = 0;
166 
167         if (!size)
168                 return 0;
169 
170         if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip))
171                 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
172                                               size);
173 
174         if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
175                 return -EINVAL;
176 
177         if (gfs2_is_stuffed(ip)) {
178                 error = gfs2_unstuff_dinode(ip, NULL);
179                 if (error)
180                         return error;
181         }
182 
183         lblock = offset;
184         o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
185 
186         while (copied < size) {
187                 unsigned int amount;
188                 struct buffer_head *bh;
189 
190                 amount = size - copied;
191                 if (amount > sdp->sd_sb.sb_bsize - o)
192                         amount = sdp->sd_sb.sb_bsize - o;
193 
194                 if (!extlen) {
195                         new = 1;
196                         error = gfs2_extent_map(&ip->i_inode, lblock, &new,
197                                                 &dblock, &extlen);
198                         if (error)
199                                 goto fail;
200                         error = -EIO;
201                         if (gfs2_assert_withdraw(sdp, dblock))
202                                 goto fail;
203                 }
204 
205                 if (amount == sdp->sd_jbsize || new)
206                         error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
207                 else
208                         error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
209 
210                 if (error)
211                         goto fail;
212 
213                 gfs2_trans_add_meta(ip->i_gl, bh);
214                 memcpy(bh->b_data + o, buf, amount);
215                 brelse(bh);
216 
217                 buf += amount;
218                 copied += amount;
219                 lblock++;
220                 dblock++;
221                 extlen--;
222 
223                 o = sizeof(struct gfs2_meta_header);
224         }
225 
226 out:
227         error = gfs2_meta_inode_buffer(ip, &dibh);
228         if (error)
229                 return error;
230 
231         if (ip->i_inode.i_size < offset + copied)
232                 i_size_write(&ip->i_inode, offset + copied);
233         ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
234 
235         gfs2_trans_add_meta(ip->i_gl, dibh);
236         gfs2_dinode_out(ip, dibh->b_data);
237         brelse(dibh);
238 
239         return copied;
240 fail:
241         if (copied)
242                 goto out;
243         return error;
244 }
245 
246 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
247                                  unsigned int size)
248 {
249         struct buffer_head *dibh;
250         int error;
251 
252         error = gfs2_meta_inode_buffer(ip, &dibh);
253         if (!error) {
254                 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
255                 brelse(dibh);
256         }
257 
258         return (error) ? error : size;
259 }
260 
261 
262 /**
263  * gfs2_dir_read_data - Read a data from a directory inode
264  * @ip: The GFS2 Inode
265  * @buf: The buffer to place result into
266  * @size: Amount of data to transfer
267  *
268  * Returns: The amount of data actually copied or the error
269  */
270 static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
271                               unsigned int size)
272 {
273         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
274         u64 lblock, dblock;
275         u32 extlen = 0;
276         unsigned int o;
277         int copied = 0;
278         int error = 0;
279 
280         if (gfs2_is_stuffed(ip))
281                 return gfs2_dir_read_stuffed(ip, buf, size);
282 
283         if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
284                 return -EINVAL;
285 
286         lblock = 0;
287         o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
288 
289         while (copied < size) {
290                 unsigned int amount;
291                 struct buffer_head *bh;
292                 int new;
293 
294                 amount = size - copied;
295                 if (amount > sdp->sd_sb.sb_bsize - o)
296                         amount = sdp->sd_sb.sb_bsize - o;
297 
298                 if (!extlen) {
299                         new = 0;
300                         error = gfs2_extent_map(&ip->i_inode, lblock, &new,
301                                                 &dblock, &extlen);
302                         if (error || !dblock)
303                                 goto fail;
304                         BUG_ON(extlen < 1);
305                         bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
306                 } else {
307                         error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
308                         if (error)
309                                 goto fail;
310                 }
311                 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
312                 if (error) {
313                         brelse(bh);
314                         goto fail;
315                 }
316                 dblock++;
317                 extlen--;
318                 memcpy(buf, bh->b_data + o, amount);
319                 brelse(bh);
320                 buf += (amount/sizeof(__be64));
321                 copied += amount;
322                 lblock++;
323                 o = sizeof(struct gfs2_meta_header);
324         }
325 
326         return copied;
327 fail:
328         return (copied) ? copied : error;
329 }
330 
331 /**
332  * gfs2_dir_get_hash_table - Get pointer to the dir hash table
333  * @ip: The inode in question
334  *
335  * Returns: The hash table or an error
336  */
337 
338 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
339 {
340         struct inode *inode = &ip->i_inode;
341         int ret;
342         u32 hsize;
343         __be64 *hc;
344 
345         BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
346 
347         hc = ip->i_hash_cache;
348         if (hc)
349                 return hc;
350 
351         hsize = BIT(ip->i_depth);
352         hsize *= sizeof(__be64);
353         if (hsize != i_size_read(&ip->i_inode)) {
354                 gfs2_consist_inode(ip);
355                 return ERR_PTR(-EIO);
356         }
357 
358         hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
359         if (hc == NULL)
360                 hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL);
361 
362         if (hc == NULL)
363                 return ERR_PTR(-ENOMEM);
364 
365         ret = gfs2_dir_read_data(ip, hc, hsize);
366         if (ret < 0) {
367                 kvfree(hc);
368                 return ERR_PTR(ret);
369         }
370 
371         spin_lock(&inode->i_lock);
372         if (likely(!ip->i_hash_cache)) {
373                 ip->i_hash_cache = hc;
374                 hc = NULL;
375         }
376         spin_unlock(&inode->i_lock);
377         kvfree(hc);
378 
379         return ip->i_hash_cache;
380 }
381 
382 /**
383  * gfs2_dir_hash_inval - Invalidate dir hash
384  * @ip: The directory inode
385  *
386  * Must be called with an exclusive glock, or during glock invalidation.
387  */
388 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
389 {
390         __be64 *hc;
391 
392         spin_lock(&ip->i_inode.i_lock);
393         hc = ip->i_hash_cache;
394         ip->i_hash_cache = NULL;
395         spin_unlock(&ip->i_inode.i_lock);
396 
397         kvfree(hc);
398 }
399 
400 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
401 {
402         return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
403 }
404 
405 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
406                                      const struct qstr *name, int ret)
407 {
408         if (!gfs2_dirent_sentinel(dent) &&
409             be32_to_cpu(dent->de_hash) == name->hash &&
410             be16_to_cpu(dent->de_name_len) == name->len &&
411             memcmp(dent+1, name->name, name->len) == 0)
412                 return ret;
413         return 0;
414 }
415 
416 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
417                             const struct qstr *name,
418                             void *opaque)
419 {
420         return __gfs2_dirent_find(dent, name, 1);
421 }
422 
423 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
424                             const struct qstr *name,
425                             void *opaque)
426 {
427         return __gfs2_dirent_find(dent, name, 2);
428 }
429 
430 /*
431  * name->name holds ptr to start of block.
432  * name->len holds size of block.
433  */
434 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
435                             const struct qstr *name,
436                             void *opaque)
437 {
438         const char *start = name->name;
439         const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
440         if (name->len == (end - start))
441                 return 1;
442         return 0;
443 }
444 
445 /* Look for the dirent that contains the offset specified in data. Once we
446  * find that dirent, there must be space available there for the new dirent */
447 static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
448                                   const struct qstr *name,
449                                   void *ptr)
450 {
451         unsigned required = GFS2_DIRENT_SIZE(name->len);
452         unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
453         unsigned totlen = be16_to_cpu(dent->de_rec_len);
454 
455         if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
456                 return 0;
457         if (gfs2_dirent_sentinel(dent))
458                 actual = 0;
459         if (ptr < (void *)dent + actual)
460                 return -1;
461         if ((void *)dent + totlen >= ptr + required)
462                 return 1;
463         return -1;
464 }
465 
466 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
467                                   const struct qstr *name,
468                                   void *opaque)
469 {
470         unsigned required = GFS2_DIRENT_SIZE(name->len);
471         unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
472         unsigned totlen = be16_to_cpu(dent->de_rec_len);
473 
474         if (gfs2_dirent_sentinel(dent))
475                 actual = 0;
476         if (totlen - actual >= required)
477                 return 1;
478         return 0;
479 }
480 
481 struct dirent_gather {
482         const struct gfs2_dirent **pdent;
483         unsigned offset;
484 };
485 
486 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
487                               const struct qstr *name,
488                               void *opaque)
489 {
490         struct dirent_gather *g = opaque;
491         if (!gfs2_dirent_sentinel(dent)) {
492                 g->pdent[g->offset++] = dent;
493         }
494         return 0;
495 }
496 
497 /*
498  * Other possible things to check:
499  * - Inode located within filesystem size (and on valid block)
500  * - Valid directory entry type
501  * Not sure how heavy-weight we want to make this... could also check
502  * hash is correct for example, but that would take a lot of extra time.
503  * For now the most important thing is to check that the various sizes
504  * are correct.
505  */
506 static int gfs2_check_dirent(struct gfs2_sbd *sdp,
507                              struct gfs2_dirent *dent, unsigned int offset,
508                              unsigned int size, unsigned int len, int first)
509 {
510         const char *msg = "gfs2_dirent too small";
511         if (unlikely(size < sizeof(struct gfs2_dirent)))
512                 goto error;
513         msg = "gfs2_dirent misaligned";
514         if (unlikely(offset & 0x7))
515                 goto error;
516         msg = "gfs2_dirent points beyond end of block";
517         if (unlikely(offset + size > len))
518                 goto error;
519         msg = "zero inode number";
520         if (unlikely(!first && gfs2_dirent_sentinel(dent)))
521                 goto error;
522         msg = "name length is greater than space in dirent";
523         if (!gfs2_dirent_sentinel(dent) &&
524             unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
525                      size))
526                 goto error;
527         return 0;
528 error:
529         fs_warn(sdp, "%s: %s (%s)\n",
530                 __func__, msg, first ? "first in block" : "not first in block");
531         return -EIO;
532 }
533 
534 static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf)
535 {
536         const struct gfs2_meta_header *h = buf;
537         int offset;
538 
539         BUG_ON(buf == NULL);
540 
541         switch(be32_to_cpu(h->mh_type)) {
542         case GFS2_METATYPE_LF:
543                 offset = sizeof(struct gfs2_leaf);
544                 break;
545         case GFS2_METATYPE_DI:
546                 offset = sizeof(struct gfs2_dinode);
547                 break;
548         default:
549                 goto wrong_type;
550         }
551         return offset;
552 wrong_type:
553         fs_warn(sdp, "%s: wrong block type %u\n", __func__,
554                 be32_to_cpu(h->mh_type));
555         return -1;
556 }
557 
558 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
559                                             unsigned int len, gfs2_dscan_t scan,
560                                             const struct qstr *name,
561                                             void *opaque)
562 {
563         struct gfs2_dirent *dent, *prev;
564         unsigned offset;
565         unsigned size;
566         int ret = 0;
567 
568         ret = gfs2_dirent_offset(GFS2_SB(inode), buf);
569         if (ret < 0)
570                 goto consist_inode;
571 
572         offset = ret;
573         prev = NULL;
574         dent = buf + offset;
575         size = be16_to_cpu(dent->de_rec_len);
576         if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1))
577                 goto consist_inode;
578         do {
579                 ret = scan(dent, name, opaque);
580                 if (ret)
581                         break;
582                 offset += size;
583                 if (offset == len)
584                         break;
585                 prev = dent;
586                 dent = buf + offset;
587                 size = be16_to_cpu(dent->de_rec_len);
588                 if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size,
589                                       len, 0))
590                         goto consist_inode;
591         } while(1);
592 
593         switch(ret) {
594         case 0:
595                 return NULL;
596         case 1:
597                 return dent;
598         case 2:
599                 return prev ? prev : dent;
600         default:
601                 BUG_ON(ret > 0);
602                 return ERR_PTR(ret);
603         }
604 
605 consist_inode:
606         gfs2_consist_inode(GFS2_I(inode));
607         return ERR_PTR(-EIO);
608 }
609 
610 static int dirent_check_reclen(struct gfs2_inode *dip,
611                                const struct gfs2_dirent *d, const void *end_p)
612 {
613         const void *ptr = d;
614         u16 rec_len = be16_to_cpu(d->de_rec_len);
615 
616         if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
617                 goto broken;
618         ptr += rec_len;
619         if (ptr < end_p)
620                 return rec_len;
621         if (ptr == end_p)
622                 return -ENOENT;
623 broken:
624         gfs2_consist_inode(dip);
625         return -EIO;
626 }
627 
628 /**
629  * dirent_next - Next dirent
630  * @dip: the directory
631  * @bh: The buffer
632  * @dent: Pointer to list of dirents
633  *
634  * Returns: 0 on success, error code otherwise
635  */
636 
637 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
638                        struct gfs2_dirent **dent)
639 {
640         struct gfs2_dirent *cur = *dent, *tmp;
641         char *bh_end = bh->b_data + bh->b_size;
642         int ret;
643 
644         ret = dirent_check_reclen(dip, cur, bh_end);
645         if (ret < 0)
646                 return ret;
647 
648         tmp = (void *)cur + ret;
649         ret = dirent_check_reclen(dip, tmp, bh_end);
650         if (ret == -EIO)
651                 return ret;
652 
653         /* Only the first dent could ever have de_inum.no_addr == 0 */
654         if (gfs2_dirent_sentinel(tmp)) {
655                 gfs2_consist_inode(dip);
656                 return -EIO;
657         }
658 
659         *dent = tmp;
660         return 0;
661 }
662 
663 /**
664  * dirent_del - Delete a dirent
665  * @dip: The GFS2 inode
666  * @bh: The buffer
667  * @prev: The previous dirent
668  * @cur: The current dirent
669  *
670  */
671 
672 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
673                        struct gfs2_dirent *prev, struct gfs2_dirent *cur)
674 {
675         u16 cur_rec_len, prev_rec_len;
676 
677         if (gfs2_dirent_sentinel(cur)) {
678                 gfs2_consist_inode(dip);
679                 return;
680         }
681 
682         gfs2_trans_add_meta(dip->i_gl, bh);
683 
684         /* If there is no prev entry, this is the first entry in the block.
685            The de_rec_len is already as big as it needs to be.  Just zero
686            out the inode number and return.  */
687 
688         if (!prev) {
689                 cur->de_inum.no_addr = 0;
690                 cur->de_inum.no_formal_ino = 0;
691                 return;
692         }
693 
694         /*  Combine this dentry with the previous one.  */
695 
696         prev_rec_len = be16_to_cpu(prev->de_rec_len);
697         cur_rec_len = be16_to_cpu(cur->de_rec_len);
698 
699         if ((char *)prev + prev_rec_len != (char *)cur)
700                 gfs2_consist_inode(dip);
701         if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
702                 gfs2_consist_inode(dip);
703 
704         prev_rec_len += cur_rec_len;
705         prev->de_rec_len = cpu_to_be16(prev_rec_len);
706 }
707 
708 
709 static struct gfs2_dirent *do_init_dirent(struct inode *inode,
710                                           struct gfs2_dirent *dent,
711                                           const struct qstr *name,
712                                           struct buffer_head *bh,
713                                           unsigned offset)
714 {
715         struct gfs2_inode *ip = GFS2_I(inode);
716         struct gfs2_dirent *ndent;
717         unsigned totlen;
718 
719         totlen = be16_to_cpu(dent->de_rec_len);
720         BUG_ON(offset + name->len > totlen);
721         gfs2_trans_add_meta(ip->i_gl, bh);
722         ndent = (struct gfs2_dirent *)((char *)dent + offset);
723         dent->de_rec_len = cpu_to_be16(offset);
724         gfs2_qstr2dirent(name, totlen - offset, ndent);
725         return ndent;
726 }
727 
728 
729 /*
730  * Takes a dent from which to grab space as an argument. Returns the
731  * newly created dent.
732  */
733 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
734                                             struct gfs2_dirent *dent,
735                                             const struct qstr *name,
736                                             struct buffer_head *bh)
737 {
738         unsigned offset = 0;
739 
740         if (!gfs2_dirent_sentinel(dent))
741                 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
742         return do_init_dirent(inode, dent, name, bh, offset);
743 }
744 
745 static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
746                                                    struct buffer_head *bh,
747                                                    const struct qstr *name,
748                                                    void *ptr)
749 {
750         struct gfs2_dirent *dent;
751         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
752                                 gfs2_dirent_find_offset, name, ptr);
753         if (IS_ERR_OR_NULL(dent))
754                 return dent;
755         return do_init_dirent(inode, dent, name, bh,
756                               (unsigned)(ptr - (void *)dent));
757 }
758 
759 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
760                     struct buffer_head **bhp)
761 {
762         int error;
763 
764         error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
765         if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
766                 /* pr_info("block num=%llu\n", leaf_no); */
767                 error = -EIO;
768         }
769 
770         return error;
771 }
772 
773 /**
774  * get_leaf_nr - Get a leaf number associated with the index
775  * @dip: The GFS2 inode
776  * @index:
777  * @leaf_out:
778  *
779  * Returns: 0 on success, error code otherwise
780  */
781 
782 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
783                        u64 *leaf_out)
784 {
785         __be64 *hash;
786         int error;
787 
788         hash = gfs2_dir_get_hash_table(dip);
789         error = PTR_ERR_OR_ZERO(hash);
790 
791         if (!error)
792                 *leaf_out = be64_to_cpu(*(hash + index));
793 
794         return error;
795 }
796 
797 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
798                           struct buffer_head **bh_out)
799 {
800         u64 leaf_no;
801         int error;
802 
803         error = get_leaf_nr(dip, index, &leaf_no);
804         if (!error)
805                 error = get_leaf(dip, leaf_no, bh_out);
806 
807         return error;
808 }
809 
810 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
811                                               const struct qstr *name,
812                                               gfs2_dscan_t scan,
813                                               struct buffer_head **pbh)
814 {
815         struct buffer_head *bh;
816         struct gfs2_dirent *dent;
817         struct gfs2_inode *ip = GFS2_I(inode);
818         int error;
819 
820         if (ip->i_diskflags & GFS2_DIF_EXHASH) {
821                 struct gfs2_leaf *leaf;
822                 unsigned int hsize = BIT(ip->i_depth);
823                 unsigned int index;
824                 u64 ln;
825                 if (hsize * sizeof(u64) != i_size_read(inode)) {
826                         gfs2_consist_inode(ip);
827                         return ERR_PTR(-EIO);
828                 }
829 
830                 index = name->hash >> (32 - ip->i_depth);
831                 error = get_first_leaf(ip, index, &bh);
832                 if (error)
833                         return ERR_PTR(error);
834                 do {
835                         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
836                                                 scan, name, NULL);
837                         if (dent)
838                                 goto got_dent;
839                         leaf = (struct gfs2_leaf *)bh->b_data;
840                         ln = be64_to_cpu(leaf->lf_next);
841                         brelse(bh);
842                         if (!ln)
843                                 break;
844 
845                         error = get_leaf(ip, ln, &bh);
846                 } while(!error);
847 
848                 return error ? ERR_PTR(error) : NULL;
849         }
850 
851 
852         error = gfs2_meta_inode_buffer(ip, &bh);
853         if (error)
854                 return ERR_PTR(error);
855         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
856 got_dent:
857         if (IS_ERR_OR_NULL(dent)) {
858                 brelse(bh);
859                 bh = NULL;
860         }
861         *pbh = bh;
862         return dent;
863 }
864 
865 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
866 {
867         struct gfs2_inode *ip = GFS2_I(inode);
868         unsigned int n = 1;
869         u64 bn;
870         int error;
871         struct buffer_head *bh;
872         struct gfs2_leaf *leaf;
873         struct gfs2_dirent *dent;
874         struct timespec64 tv = current_time(inode);
875 
876         error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
877         if (error)
878                 return NULL;
879         bh = gfs2_meta_new(ip->i_gl, bn);
880         if (!bh)
881                 return NULL;
882 
883         gfs2_trans_remove_revoke(GFS2_SB(inode), bn, 1);
884         gfs2_trans_add_meta(ip->i_gl, bh);
885         gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
886         leaf = (struct gfs2_leaf *)bh->b_data;
887         leaf->lf_depth = cpu_to_be16(depth);
888         leaf->lf_entries = 0;
889         leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
890         leaf->lf_next = 0;
891         leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
892         leaf->lf_dist = cpu_to_be32(1);
893         leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
894         leaf->lf_sec = cpu_to_be64(tv.tv_sec);
895         memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
896         dent = (struct gfs2_dirent *)(leaf+1);
897         gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
898         *pbh = bh;
899         return leaf;
900 }
901 
902 /**
903  * dir_make_exhash - Convert a stuffed directory into an ExHash directory
904  * @dip: The GFS2 inode
905  *
906  * Returns: 0 on success, error code otherwise
907  */
908 
909 static int dir_make_exhash(struct inode *inode)
910 {
911         struct gfs2_inode *dip = GFS2_I(inode);
912         struct gfs2_sbd *sdp = GFS2_SB(inode);
913         struct gfs2_dirent *dent;
914         struct qstr args;
915         struct buffer_head *bh, *dibh;
916         struct gfs2_leaf *leaf;
917         int y;
918         u32 x;
919         __be64 *lp;
920         u64 bn;
921         int error;
922 
923         error = gfs2_meta_inode_buffer(dip, &dibh);
924         if (error)
925                 return error;
926 
927         /*  Turn over a new leaf  */
928 
929         leaf = new_leaf(inode, &bh, 0);
930         if (!leaf)
931                 return -ENOSPC;
932         bn = bh->b_blocknr;
933 
934         gfs2_assert(sdp, dip->i_entries < BIT(16));
935         leaf->lf_entries = cpu_to_be16(dip->i_entries);
936 
937         /*  Copy dirents  */
938 
939         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
940                              sizeof(struct gfs2_dinode));
941 
942         /*  Find last entry  */
943 
944         x = 0;
945         args.len = bh->b_size - sizeof(struct gfs2_dinode) +
946                    sizeof(struct gfs2_leaf);
947         args.name = bh->b_data;
948         dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
949                                 gfs2_dirent_last, &args, NULL);
950         if (!dent) {
951                 brelse(bh);
952                 brelse(dibh);
953                 return -EIO;
954         }
955         if (IS_ERR(dent)) {
956                 brelse(bh);
957                 brelse(dibh);
958                 return PTR_ERR(dent);
959         }
960 
961         /*  Adjust the last dirent's record length
962            (Remember that dent still points to the last entry.)  */
963 
964         dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
965                 sizeof(struct gfs2_dinode) -
966                 sizeof(struct gfs2_leaf));
967 
968         brelse(bh);
969 
970         /*  We're done with the new leaf block, now setup the new
971             hash table.  */
972 
973         gfs2_trans_add_meta(dip->i_gl, dibh);
974         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
975 
976         lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
977 
978         for (x = sdp->sd_hash_ptrs; x--; lp++)
979                 *lp = cpu_to_be64(bn);
980 
981         i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
982         gfs2_add_inode_blocks(&dip->i_inode, 1);
983         dip->i_diskflags |= GFS2_DIF_EXHASH;
984 
985         for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
986         dip->i_depth = y;
987 
988         gfs2_dinode_out(dip, dibh->b_data);
989 
990         brelse(dibh);
991 
992         return 0;
993 }
994 
995 /**
996  * dir_split_leaf - Split a leaf block into two
997  * @dip: The GFS2 inode
998  * @index:
999  * @leaf_no:
1000  *
1001  * Returns: 0 on success, error code on failure
1002  */
1003 
1004 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
1005 {
1006         struct gfs2_inode *dip = GFS2_I(inode);
1007         struct buffer_head *nbh, *obh, *dibh;
1008         struct gfs2_leaf *nleaf, *oleaf;
1009         struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1010         u32 start, len, half_len, divider;
1011         u64 bn, leaf_no;
1012         __be64 *lp;
1013         u32 index;
1014         int x;
1015         int error;
1016 
1017         index = name->hash >> (32 - dip->i_depth);
1018         error = get_leaf_nr(dip, index, &leaf_no);
1019         if (error)
1020                 return error;
1021 
1022         /*  Get the old leaf block  */
1023         error = get_leaf(dip, leaf_no, &obh);
1024         if (error)
1025                 return error;
1026 
1027         oleaf = (struct gfs2_leaf *)obh->b_data;
1028         if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1029                 brelse(obh);
1030                 return 1; /* can't split */
1031         }
1032 
1033         gfs2_trans_add_meta(dip->i_gl, obh);
1034 
1035         nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1036         if (!nleaf) {
1037                 brelse(obh);
1038                 return -ENOSPC;
1039         }
1040         bn = nbh->b_blocknr;
1041 
1042         /*  Compute the start and len of leaf pointers in the hash table.  */
1043         len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1044         half_len = len >> 1;
1045         if (!half_len) {
1046                 fs_warn(GFS2_SB(inode), "i_depth %u lf_depth %u index %u\n",
1047                         dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1048                 gfs2_consist_inode(dip);
1049                 error = -EIO;
1050                 goto fail_brelse;
1051         }
1052 
1053         start = (index & ~(len - 1));
1054 
1055         /* Change the pointers.
1056            Don't bother distinguishing stuffed from non-stuffed.
1057            This code is complicated enough already. */
1058         lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS);
1059         if (!lp) {
1060                 error = -ENOMEM;
1061                 goto fail_brelse;
1062         }
1063 
1064         /*  Change the pointers  */
1065         for (x = 0; x < half_len; x++)
1066                 lp[x] = cpu_to_be64(bn);
1067 
1068         gfs2_dir_hash_inval(dip);
1069 
1070         error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1071                                     half_len * sizeof(u64));
1072         if (error != half_len * sizeof(u64)) {
1073                 if (error >= 0)
1074                         error = -EIO;
1075                 goto fail_lpfree;
1076         }
1077 
1078         kfree(lp);
1079 
1080         /*  Compute the divider  */
1081         divider = (start + half_len) << (32 - dip->i_depth);
1082 
1083         /*  Copy the entries  */
1084         dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1085 
1086         do {
1087                 next = dent;
1088                 if (dirent_next(dip, obh, &next))
1089                         next = NULL;
1090 
1091                 if (!gfs2_dirent_sentinel(dent) &&
1092                     be32_to_cpu(dent->de_hash) < divider) {
1093                         struct qstr str;
1094                         void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1095                         str.name = (char*)(dent+1);
1096                         str.len = be16_to_cpu(dent->de_name_len);
1097                         str.hash = be32_to_cpu(dent->de_hash);
1098                         new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1099                         if (IS_ERR(new)) {
1100                                 error = PTR_ERR(new);
1101                                 break;
1102                         }
1103 
1104                         new->de_inum = dent->de_inum; /* No endian worries */
1105                         new->de_type = dent->de_type; /* No endian worries */
1106                         be16_add_cpu(&nleaf->lf_entries, 1);
1107 
1108                         dirent_del(dip, obh, prev, dent);
1109 
1110                         if (!oleaf->lf_entries)
1111                                 gfs2_consist_inode(dip);
1112                         be16_add_cpu(&oleaf->lf_entries, -1);
1113 
1114                         if (!prev)
1115                                 prev = dent;
1116                 } else {
1117                         prev = dent;
1118                 }
1119                 dent = next;
1120         } while (dent);
1121 
1122         oleaf->lf_depth = nleaf->lf_depth;
1123 
1124         error = gfs2_meta_inode_buffer(dip, &dibh);
1125         if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1126                 gfs2_trans_add_meta(dip->i_gl, dibh);
1127                 gfs2_add_inode_blocks(&dip->i_inode, 1);
1128                 gfs2_dinode_out(dip, dibh->b_data);
1129                 brelse(dibh);
1130         }
1131 
1132         brelse(obh);
1133         brelse(nbh);
1134 
1135         return error;
1136 
1137 fail_lpfree:
1138         kfree(lp);
1139 
1140 fail_brelse:
1141         brelse(obh);
1142         brelse(nbh);
1143         return error;
1144 }
1145 
1146 /**
1147  * dir_double_exhash - Double size of ExHash table
1148  * @dip: The GFS2 dinode
1149  *
1150  * Returns: 0 on success, error code on failure
1151  */
1152 
1153 static int dir_double_exhash(struct gfs2_inode *dip)
1154 {
1155         struct buffer_head *dibh;
1156         u32 hsize;
1157         u32 hsize_bytes;
1158         __be64 *hc;
1159         __be64 *hc2, *h;
1160         int x;
1161         int error = 0;
1162 
1163         hsize = BIT(dip->i_depth);
1164         hsize_bytes = hsize * sizeof(__be64);
1165 
1166         hc = gfs2_dir_get_hash_table(dip);
1167         if (IS_ERR(hc))
1168                 return PTR_ERR(hc);
1169 
1170         hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN);
1171         if (hc2 == NULL)
1172                 hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL);
1173 
1174         if (!hc2)
1175                 return -ENOMEM;
1176 
1177         h = hc2;
1178         error = gfs2_meta_inode_buffer(dip, &dibh);
1179         if (error)
1180                 goto out_kfree;
1181 
1182         for (x = 0; x < hsize; x++) {
1183                 *h++ = *hc;
1184                 *h++ = *hc;
1185                 hc++;
1186         }
1187 
1188         error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1189         if (error != (hsize_bytes * 2))
1190                 goto fail;
1191 
1192         gfs2_dir_hash_inval(dip);
1193         dip->i_hash_cache = hc2;
1194         dip->i_depth++;
1195         gfs2_dinode_out(dip, dibh->b_data);
1196         brelse(dibh);
1197         return 0;
1198 
1199 fail:
1200         /* Replace original hash table & size */
1201         gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1202         i_size_write(&dip->i_inode, hsize_bytes);
1203         gfs2_dinode_out(dip, dibh->b_data);
1204         brelse(dibh);
1205 out_kfree:
1206         kvfree(hc2);
1207         return error;
1208 }
1209 
1210 /**
1211  * compare_dents - compare directory entries by hash value
1212  * @a: first dent
1213  * @b: second dent
1214  *
1215  * When comparing the hash entries of @a to @b:
1216  *   gt: returns 1
1217  *   lt: returns -1
1218  *   eq: returns 0
1219  */
1220 
1221 static int compare_dents(const void *a, const void *b)
1222 {
1223         const struct gfs2_dirent *dent_a, *dent_b;
1224         u32 hash_a, hash_b;
1225         int ret = 0;
1226 
1227         dent_a = *(const struct gfs2_dirent **)a;
1228         hash_a = dent_a->de_cookie;
1229 
1230         dent_b = *(const struct gfs2_dirent **)b;
1231         hash_b = dent_b->de_cookie;
1232 
1233         if (hash_a > hash_b)
1234                 ret = 1;
1235         else if (hash_a < hash_b)
1236                 ret = -1;
1237         else {
1238                 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1239                 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1240 
1241                 if (len_a > len_b)
1242                         ret = 1;
1243                 else if (len_a < len_b)
1244                         ret = -1;
1245                 else
1246                         ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1247         }
1248 
1249         return ret;
1250 }
1251 
1252 /**
1253  * do_filldir_main - read out directory entries
1254  * @dip: The GFS2 inode
1255  * @ctx: what to feed the entries to
1256  * @darr: an array of struct gfs2_dirent pointers to read
1257  * @entries: the number of entries in darr
1258  * @copied: pointer to int that's non-zero if a entry has been copied out
1259  *
1260  * Jump through some hoops to make sure that if there are hash collsions,
1261  * they are read out at the beginning of a buffer.  We want to minimize
1262  * the possibility that they will fall into different readdir buffers or
1263  * that someone will want to seek to that location.
1264  *
1265  * Returns: errno, >0 if the actor tells you to stop
1266  */
1267 
1268 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1269                            struct gfs2_dirent **darr, u32 entries,
1270                            u32 sort_start, int *copied)
1271 {
1272         const struct gfs2_dirent *dent, *dent_next;
1273         u64 off, off_next;
1274         unsigned int x, y;
1275         int run = 0;
1276 
1277         if (sort_start < entries)
1278                 sort(&darr[sort_start], entries - sort_start,
1279                      sizeof(struct gfs2_dirent *), compare_dents, NULL);
1280 
1281         dent_next = darr[0];
1282         off_next = dent_next->de_cookie;
1283 
1284         for (x = 0, y = 1; x < entries; x++, y++) {
1285                 dent = dent_next;
1286                 off = off_next;
1287 
1288                 if (y < entries) {
1289                         dent_next = darr[y];
1290                         off_next = dent_next->de_cookie;
1291 
1292                         if (off < ctx->pos)
1293                                 continue;
1294                         ctx->pos = off;
1295 
1296                         if (off_next == off) {
1297                                 if (*copied && !run)
1298                                         return 1;
1299                                 run = 1;
1300                         } else
1301                                 run = 0;
1302                 } else {
1303                         if (off < ctx->pos)
1304                                 continue;
1305                         ctx->pos = off;
1306                 }
1307 
1308                 if (!dir_emit(ctx, (const char *)(dent + 1),
1309                                 be16_to_cpu(dent->de_name_len),
1310                                 be64_to_cpu(dent->de_inum.no_addr),
1311                                 be16_to_cpu(dent->de_type)))
1312                         return 1;
1313 
1314                 *copied = 1;
1315         }
1316 
1317         /* Increment the ctx->pos by one, so the next time we come into the
1318            do_filldir fxn, we get the next entry instead of the last one in the
1319            current leaf */
1320 
1321         ctx->pos++;
1322 
1323         return 0;
1324 }
1325 
1326 static void *gfs2_alloc_sort_buffer(unsigned size)
1327 {
1328         void *ptr = NULL;
1329 
1330         if (size < KMALLOC_MAX_SIZE)
1331                 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1332         if (!ptr)
1333                 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1334         return ptr;
1335 }
1336 
1337 
1338 static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1339                             unsigned leaf_nr, struct gfs2_dirent **darr,
1340                             unsigned entries)
1341 {
1342         int sort_id = -1;
1343         int i;
1344         
1345         for (i = 0; i < entries; i++) {
1346                 unsigned offset;
1347 
1348                 darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1349                 darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1350 
1351                 if (!sdp->sd_args.ar_loccookie)
1352                         continue;
1353                 offset = (char *)(darr[i]) -
1354                         (bh->b_data + gfs2_dirent_offset(sdp, bh->b_data));
1355                 offset /= GFS2_MIN_DIRENT_SIZE;
1356                 offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1357                 if (offset >= GFS2_USE_HASH_FLAG ||
1358                     leaf_nr >= GFS2_USE_HASH_FLAG) {
1359                         darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1360                         if (sort_id < 0)
1361                                 sort_id = i;
1362                         continue;
1363                 }
1364                 darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1365                 darr[i]->de_cookie |= offset;
1366         }
1367         return sort_id;
1368 }       
1369 
1370 
1371 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1372                               int *copied, unsigned *depth,
1373                               u64 leaf_no)
1374 {
1375         struct gfs2_inode *ip = GFS2_I(inode);
1376         struct gfs2_sbd *sdp = GFS2_SB(inode);
1377         struct buffer_head *bh;
1378         struct gfs2_leaf *lf;
1379         unsigned entries = 0, entries2 = 0;
1380         unsigned leaves = 0, leaf = 0, offset, sort_offset;
1381         struct gfs2_dirent **darr, *dent;
1382         struct dirent_gather g;
1383         struct buffer_head **larr;
1384         int error, i, need_sort = 0, sort_id;
1385         u64 lfn = leaf_no;
1386 
1387         do {
1388                 error = get_leaf(ip, lfn, &bh);
1389                 if (error)
1390                         goto out;
1391                 lf = (struct gfs2_leaf *)bh->b_data;
1392                 if (leaves == 0)
1393                         *depth = be16_to_cpu(lf->lf_depth);
1394                 entries += be16_to_cpu(lf->lf_entries);
1395                 leaves++;
1396                 lfn = be64_to_cpu(lf->lf_next);
1397                 brelse(bh);
1398         } while(lfn);
1399 
1400         if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1401                 need_sort = 1;
1402                 sort_offset = 0;
1403         }
1404 
1405         if (!entries)
1406                 return 0;
1407 
1408         error = -ENOMEM;
1409         /*
1410          * The extra 99 entries are not normally used, but are a buffer
1411          * zone in case the number of entries in the leaf is corrupt.
1412          * 99 is the maximum number of entries that can fit in a single
1413          * leaf block.
1414          */
1415         larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1416         if (!larr)
1417                 goto out;
1418         darr = (struct gfs2_dirent **)(larr + leaves);
1419         g.pdent = (const struct gfs2_dirent **)darr;
1420         g.offset = 0;
1421         lfn = leaf_no;
1422 
1423         do {
1424                 error = get_leaf(ip, lfn, &bh);
1425                 if (error)
1426                         goto out_free;
1427                 lf = (struct gfs2_leaf *)bh->b_data;
1428                 lfn = be64_to_cpu(lf->lf_next);
1429                 if (lf->lf_entries) {
1430                         offset = g.offset;
1431                         entries2 += be16_to_cpu(lf->lf_entries);
1432                         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1433                                                 gfs2_dirent_gather, NULL, &g);
1434                         error = PTR_ERR(dent);
1435                         if (IS_ERR(dent))
1436                                 goto out_free;
1437                         if (entries2 != g.offset) {
1438                                 fs_warn(sdp, "Number of entries corrupt in dir "
1439                                                 "leaf %llu, entries2 (%u) != "
1440                                                 "g.offset (%u)\n",
1441                                         (unsigned long long)bh->b_blocknr,
1442                                         entries2, g.offset);
1443                                 gfs2_consist_inode(ip);
1444                                 error = -EIO;
1445                                 goto out_free;
1446                         }
1447                         error = 0;
1448                         sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1449                                                    be16_to_cpu(lf->lf_entries));
1450                         if (!need_sort && sort_id >= 0) {
1451                                 need_sort = 1;
1452                                 sort_offset = offset + sort_id;
1453                         }
1454                         larr[leaf++] = bh;
1455                 } else {
1456                         larr[leaf++] = NULL;
1457                         brelse(bh);
1458                 }
1459         } while(lfn);
1460 
1461         BUG_ON(entries2 != entries);
1462         error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1463                                 sort_offset : entries, copied);
1464 out_free:
1465         for(i = 0; i < leaf; i++)
1466                 if (larr[i])
1467                         brelse(larr[i]);
1468         kvfree(larr);
1469 out:
1470         return error;
1471 }
1472 
1473 /**
1474  * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1475  *
1476  * Note: we can't calculate each index like dir_e_read can because we don't
1477  * have the leaf, and therefore we don't have the depth, and therefore we
1478  * don't have the length. So we have to just read enough ahead to make up
1479  * for the loss of information.
1480  */
1481 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1482                                struct file_ra_state *f_ra)
1483 {
1484         struct gfs2_inode *ip = GFS2_I(inode);
1485         struct gfs2_glock *gl = ip->i_gl;
1486         struct buffer_head *bh;
1487         u64 blocknr = 0, last;
1488         unsigned count;
1489 
1490         /* First check if we've already read-ahead for the whole range. */
1491         if (index + MAX_RA_BLOCKS < f_ra->start)
1492                 return;
1493 
1494         f_ra->start = max((pgoff_t)index, f_ra->start);
1495         for (count = 0; count < MAX_RA_BLOCKS; count++) {
1496                 if (f_ra->start >= hsize) /* if exceeded the hash table */
1497                         break;
1498 
1499                 last = blocknr;
1500                 blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1501                 f_ra->start++;
1502                 if (blocknr == last)
1503                         continue;
1504 
1505                 bh = gfs2_getbuf(gl, blocknr, 1);
1506                 if (trylock_buffer(bh)) {
1507                         if (buffer_uptodate(bh)) {
1508                                 unlock_buffer(bh);
1509                                 brelse(bh);
1510                                 continue;
1511                         }
1512                         bh->b_end_io = end_buffer_read_sync;
1513                         submit_bh(REQ_OP_READ,
1514                                   REQ_RAHEAD | REQ_META | REQ_PRIO,
1515                                   bh);
1516                         continue;
1517                 }
1518                 brelse(bh);
1519         }
1520 }
1521 
1522 /**
1523  * dir_e_read - Reads the entries from a directory into a filldir buffer
1524  * @dip: dinode pointer
1525  * @ctx: actor to feed the entries to
1526  *
1527  * Returns: errno
1528  */
1529 
1530 static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1531                       struct file_ra_state *f_ra)
1532 {
1533         struct gfs2_inode *dip = GFS2_I(inode);
1534         u32 hsize, len = 0;
1535         u32 hash, index;
1536         __be64 *lp;
1537         int copied = 0;
1538         int error = 0;
1539         unsigned depth = 0;
1540 
1541         hsize = BIT(dip->i_depth);
1542         hash = gfs2_dir_offset2hash(ctx->pos);
1543         index = hash >> (32 - dip->i_depth);
1544 
1545         if (dip->i_hash_cache == NULL)
1546                 f_ra->start = 0;
1547         lp = gfs2_dir_get_hash_table(dip);
1548         if (IS_ERR(lp))
1549                 return PTR_ERR(lp);
1550 
1551         gfs2_dir_readahead(inode, hsize, index, f_ra);
1552 
1553         while (index < hsize) {
1554                 error = gfs2_dir_read_leaf(inode, ctx,
1555                                            &copied, &depth,
1556                                            be64_to_cpu(lp[index]));
1557                 if (error)
1558                         break;
1559 
1560                 len = BIT(dip->i_depth - depth);
1561                 index = (index & ~(len - 1)) + len;
1562         }
1563 
1564         if (error > 0)
1565                 error = 0;
1566         return error;
1567 }
1568 
1569 int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1570                   struct file_ra_state *f_ra)
1571 {
1572         struct gfs2_inode *dip = GFS2_I(inode);
1573         struct gfs2_sbd *sdp = GFS2_SB(inode);
1574         struct dirent_gather g;
1575         struct gfs2_dirent **darr, *dent;
1576         struct buffer_head *dibh;
1577         int copied = 0;
1578         int error;
1579 
1580         if (!dip->i_entries)
1581                 return 0;
1582 
1583         if (dip->i_diskflags & GFS2_DIF_EXHASH)
1584                 return dir_e_read(inode, ctx, f_ra);
1585 
1586         if (!gfs2_is_stuffed(dip)) {
1587                 gfs2_consist_inode(dip);
1588                 return -EIO;
1589         }
1590 
1591         error = gfs2_meta_inode_buffer(dip, &dibh);
1592         if (error)
1593                 return error;
1594 
1595         error = -ENOMEM;
1596         /* 96 is max number of dirents which can be stuffed into an inode */
1597         darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS);
1598         if (darr) {
1599                 g.pdent = (const struct gfs2_dirent **)darr;
1600                 g.offset = 0;
1601                 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1602                                         gfs2_dirent_gather, NULL, &g);
1603                 if (IS_ERR(dent)) {
1604                         error = PTR_ERR(dent);
1605                         goto out;
1606                 }
1607                 if (dip->i_entries != g.offset) {
1608                         fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1609                                 "ip->i_entries (%u) != g.offset (%u)\n",
1610                                 (unsigned long long)dip->i_no_addr,
1611                                 dip->i_entries,
1612                                 g.offset);
1613                         gfs2_consist_inode(dip);
1614                         error = -EIO;
1615                         goto out;
1616                 }
1617                 gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1618                 error = do_filldir_main(dip, ctx, darr,
1619                                         dip->i_entries, 0, &copied);
1620 out:
1621                 kfree(darr);
1622         }
1623 
1624         if (error > 0)
1625                 error = 0;
1626 
1627         brelse(dibh);
1628 
1629         return error;
1630 }
1631 
1632 /**
1633  * gfs2_dir_search - Search a directory
1634  * @dip: The GFS2 dir inode
1635  * @name: The name we are looking up
1636  * @fail_on_exist: Fail if the name exists rather than looking it up
1637  *
1638  * This routine searches a directory for a file or another directory.
1639  * Assumes a glock is held on dip.
1640  *
1641  * Returns: errno
1642  */
1643 
1644 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1645                               bool fail_on_exist)
1646 {
1647         struct buffer_head *bh;
1648         struct gfs2_dirent *dent;
1649         u64 addr, formal_ino;
1650         u16 dtype;
1651 
1652         dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1653         if (dent) {
1654                 struct inode *inode;
1655                 u16 rahead;
1656 
1657                 if (IS_ERR(dent))
1658                         return ERR_CAST(dent);
1659                 dtype = be16_to_cpu(dent->de_type);
1660                 rahead = be16_to_cpu(dent->de_rahead);
1661                 addr = be64_to_cpu(dent->de_inum.no_addr);
1662                 formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1663                 brelse(bh);
1664                 if (fail_on_exist)
1665                         return ERR_PTR(-EEXIST);
1666                 inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
1667                                           GFS2_BLKST_FREE /* ignore */);
1668                 if (!IS_ERR(inode))
1669                         GFS2_I(inode)->i_rahead = rahead;
1670                 return inode;
1671         }
1672         return ERR_PTR(-ENOENT);
1673 }
1674 
1675 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1676                    const struct gfs2_inode *ip)
1677 {
1678         struct buffer_head *bh;
1679         struct gfs2_dirent *dent;
1680         int ret = -ENOENT;
1681 
1682         dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1683         if (dent) {
1684                 if (IS_ERR(dent))
1685                         return PTR_ERR(dent);
1686                 if (ip) {
1687                         if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1688                                 goto out;
1689                         if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1690                             ip->i_no_formal_ino)
1691                                 goto out;
1692                         if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1693                             be16_to_cpu(dent->de_type))) {
1694                                 gfs2_consist_inode(GFS2_I(dir));
1695                                 ret = -EIO;
1696                                 goto out;
1697                         }
1698                 }
1699                 ret = 0;
1700 out:
1701                 brelse(bh);
1702         }
1703         return ret;
1704 }
1705 
1706 /**
1707  * dir_new_leaf - Add a new leaf onto hash chain
1708  * @inode: The directory
1709  * @name: The name we are adding
1710  *
1711  * This adds a new dir leaf onto an existing leaf when there is not
1712  * enough space to add a new dir entry. This is a last resort after
1713  * we've expanded the hash table to max size and also split existing
1714  * leaf blocks, so it will only occur for very large directories.
1715  *
1716  * The dist parameter is set to 1 for leaf blocks directly attached
1717  * to the hash table, 2 for one layer of indirection, 3 for two layers
1718  * etc. We are thus able to tell the difference between an old leaf
1719  * with dist set to zero (i.e. "don't know") and a new one where we
1720  * set this information for debug/fsck purposes.
1721  *
1722  * Returns: 0 on success, or -ve on error
1723  */
1724 
1725 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1726 {
1727         struct buffer_head *bh, *obh;
1728         struct gfs2_inode *ip = GFS2_I(inode);
1729         struct gfs2_leaf *leaf, *oleaf;
1730         u32 dist = 1;
1731         int error;
1732         u32 index;
1733         u64 bn;
1734 
1735         index = name->hash >> (32 - ip->i_depth);
1736         error = get_first_leaf(ip, index, &obh);
1737         if (error)
1738                 return error;
1739         do {
1740                 dist++;
1741                 oleaf = (struct gfs2_leaf *)obh->b_data;
1742                 bn = be64_to_cpu(oleaf->lf_next);
1743                 if (!bn)
1744                         break;
1745                 brelse(obh);
1746                 error = get_leaf(ip, bn, &obh);
1747                 if (error)
1748                         return error;
1749         } while(1);
1750 
1751         gfs2_trans_add_meta(ip->i_gl, obh);
1752 
1753         leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1754         if (!leaf) {
1755                 brelse(obh);
1756                 return -ENOSPC;
1757         }
1758         leaf->lf_dist = cpu_to_be32(dist);
1759         oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1760         brelse(bh);
1761         brelse(obh);
1762 
1763         error = gfs2_meta_inode_buffer(ip, &bh);
1764         if (error)
1765                 return error;
1766         gfs2_trans_add_meta(ip->i_gl, bh);
1767         gfs2_add_inode_blocks(&ip->i_inode, 1);
1768         gfs2_dinode_out(ip, bh->b_data);
1769         brelse(bh);
1770         return 0;
1771 }
1772 
1773 static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
1774 {
1775         u64 where = ip->i_no_addr + 1;
1776         if (ip->i_eattr == where)
1777                 return 1;
1778         return 0;
1779 }
1780 
1781 /**
1782  * gfs2_dir_add - Add new filename into directory
1783  * @inode: The directory inode
1784  * @name: The new name
1785  * @nip: The GFS2 inode to be linked in to the directory
1786  * @da: The directory addition info
1787  *
1788  * If the call to gfs2_diradd_alloc_required resulted in there being
1789  * no need to allocate any new directory blocks, then it will contain
1790  * a pointer to the directory entry and the bh in which it resides. We
1791  * can use that without having to repeat the search. If there was no
1792  * free space, then we must now create more space.
1793  *
1794  * Returns: 0 on success, error code on failure
1795  */
1796 
1797 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1798                  const struct gfs2_inode *nip, struct gfs2_diradd *da)
1799 {
1800         struct gfs2_inode *ip = GFS2_I(inode);
1801         struct buffer_head *bh = da->bh;
1802         struct gfs2_dirent *dent = da->dent;
1803         struct timespec64 tv;
1804         struct gfs2_leaf *leaf;
1805         int error;
1806 
1807         while(1) {
1808                 if (da->bh == NULL) {
1809                         dent = gfs2_dirent_search(inode, name,
1810                                                   gfs2_dirent_find_space, &bh);
1811                 }
1812                 if (dent) {
1813                         if (IS_ERR(dent))
1814                                 return PTR_ERR(dent);
1815                         dent = gfs2_init_dirent(inode, dent, name, bh);
1816                         gfs2_inum_out(nip, dent);
1817                         dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1818                         dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1819                         tv = current_time(&ip->i_inode);
1820                         if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1821                                 leaf = (struct gfs2_leaf *)bh->b_data;
1822                                 be16_add_cpu(&leaf->lf_entries, 1);
1823                                 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1824                                 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1825                         }
1826                         da->dent = NULL;
1827                         da->bh = NULL;
1828                         brelse(bh);
1829                         ip->i_entries++;
1830                         ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv;
1831                         if (S_ISDIR(nip->i_inode.i_mode))
1832                                 inc_nlink(&ip->i_inode);
1833                         mark_inode_dirty(inode);
1834                         error = 0;
1835                         break;
1836                 }
1837                 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1838                         error = dir_make_exhash(inode);
1839                         if (error)
1840                                 break;
1841                         continue;
1842                 }
1843                 error = dir_split_leaf(inode, name);
1844                 if (error == 0)
1845                         continue;
1846                 if (error < 0)
1847                         break;
1848                 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1849                         error = dir_double_exhash(ip);
1850                         if (error)
1851                                 break;
1852                         error = dir_split_leaf(inode, name);
1853                         if (error < 0)
1854                                 break;
1855                         if (error == 0)
1856                                 continue;
1857                 }
1858                 error = dir_new_leaf(inode, name);
1859                 if (!error)
1860                         continue;
1861                 error = -ENOSPC;
1862                 break;
1863         }
1864         return error;
1865 }
1866 
1867 
1868 /**
1869  * gfs2_dir_del - Delete a directory entry
1870  * @dip: The GFS2 inode
1871  * @filename: The filename
1872  *
1873  * Returns: 0 on success, error code on failure
1874  */
1875 
1876 int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1877 {
1878         const struct qstr *name = &dentry->d_name;
1879         struct gfs2_dirent *dent, *prev = NULL;
1880         struct buffer_head *bh;
1881         struct timespec64 tv = current_time(&dip->i_inode);
1882 
1883         /* Returns _either_ the entry (if its first in block) or the
1884            previous entry otherwise */
1885         dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1886         if (!dent) {
1887                 gfs2_consist_inode(dip);
1888                 return -EIO;
1889         }
1890         if (IS_ERR(dent)) {
1891                 gfs2_consist_inode(dip);
1892                 return PTR_ERR(dent);
1893         }
1894         /* If not first in block, adjust pointers accordingly */
1895         if (gfs2_dirent_find(dent, name, NULL) == 0) {
1896                 prev = dent;
1897                 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1898         }
1899 
1900         dirent_del(dip, bh, prev, dent);
1901         if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1902                 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1903                 u16 entries = be16_to_cpu(leaf->lf_entries);
1904                 if (!entries)
1905                         gfs2_consist_inode(dip);
1906                 leaf->lf_entries = cpu_to_be16(--entries);
1907                 leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1908                 leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1909         }
1910         brelse(bh);
1911 
1912         if (!dip->i_entries)
1913                 gfs2_consist_inode(dip);
1914         dip->i_entries--;
1915         dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv;
1916         if (d_is_dir(dentry))
1917                 drop_nlink(&dip->i_inode);
1918         mark_inode_dirty(&dip->i_inode);
1919 
1920         return 0;
1921 }
1922 
1923 /**
1924  * gfs2_dir_mvino - Change inode number of directory entry
1925  * @dip: The GFS2 inode
1926  * @filename:
1927  * @new_inode:
1928  *
1929  * This routine changes the inode number of a directory entry.  It's used
1930  * by rename to change ".." when a directory is moved.
1931  * Assumes a glock is held on dvp.
1932  *
1933  * Returns: errno
1934  */
1935 
1936 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1937                    const struct gfs2_inode *nip, unsigned int new_type)
1938 {
1939         struct buffer_head *bh;
1940         struct gfs2_dirent *dent;
1941 
1942         dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1943         if (!dent) {
1944                 gfs2_consist_inode(dip);
1945                 return -EIO;
1946         }
1947         if (IS_ERR(dent))
1948                 return PTR_ERR(dent);
1949 
1950         gfs2_trans_add_meta(dip->i_gl, bh);
1951         gfs2_inum_out(nip, dent);
1952         dent->de_type = cpu_to_be16(new_type);
1953         brelse(bh);
1954 
1955         dip->i_inode.i_mtime = dip->i_inode.i_ctime = current_time(&dip->i_inode);
1956         mark_inode_dirty_sync(&dip->i_inode);
1957         return 0;
1958 }
1959 
1960 /**
1961  * leaf_dealloc - Deallocate a directory leaf
1962  * @dip: the directory
1963  * @index: the hash table offset in the directory
1964  * @len: the number of pointers to this leaf
1965  * @leaf_no: the leaf number
1966  * @leaf_bh: buffer_head for the starting leaf
1967  * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1968  *
1969  * Returns: errno
1970  */
1971 
1972 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1973                         u64 leaf_no, struct buffer_head *leaf_bh,
1974                         int last_dealloc)
1975 {
1976         struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1977         struct gfs2_leaf *tmp_leaf;
1978         struct gfs2_rgrp_list rlist;
1979         struct buffer_head *bh, *dibh;
1980         u64 blk, nblk;
1981         unsigned int rg_blocks = 0, l_blocks = 0;
1982         char *ht;
1983         unsigned int x, size = len * sizeof(u64);
1984         int error;
1985 
1986         error = gfs2_rindex_update(sdp);
1987         if (error)
1988                 return error;
1989 
1990         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1991 
1992         ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1993         if (ht == NULL)
1994                 ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO,
1995                                PAGE_KERNEL);
1996         if (!ht)
1997                 return -ENOMEM;
1998 
1999         error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
2000         if (error)
2001                 goto out;
2002 
2003         /*  Count the number of leaves  */
2004         bh = leaf_bh;
2005 
2006         for (blk = leaf_no; blk; blk = nblk) {
2007                 if (blk != leaf_no) {
2008                         error = get_leaf(dip, blk, &bh);
2009                         if (error)
2010                                 goto out_rlist;
2011                 }
2012                 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2013                 nblk = be64_to_cpu(tmp_leaf->lf_next);
2014                 if (blk != leaf_no)
2015                         brelse(bh);
2016 
2017                 gfs2_rlist_add(dip, &rlist, blk);
2018                 l_blocks++;
2019         }
2020 
2021         gfs2_rlist_alloc(&rlist);
2022 
2023         for (x = 0; x < rlist.rl_rgrps; x++) {
2024                 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
2025 
2026                 rg_blocks += rgd->rd_length;
2027         }
2028 
2029         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2030         if (error)
2031                 goto out_rlist;
2032 
2033         error = gfs2_trans_begin(sdp,
2034                         rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2035                         RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
2036         if (error)
2037                 goto out_rg_gunlock;
2038 
2039         bh = leaf_bh;
2040 
2041         for (blk = leaf_no; blk; blk = nblk) {
2042                 struct gfs2_rgrpd *rgd;
2043 
2044                 if (blk != leaf_no) {
2045                         error = get_leaf(dip, blk, &bh);
2046                         if (error)
2047                                 goto out_end_trans;
2048                 }
2049                 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2050                 nblk = be64_to_cpu(tmp_leaf->lf_next);
2051                 if (blk != leaf_no)
2052                         brelse(bh);
2053 
2054                 rgd = gfs2_blk2rgrpd(sdp, blk, true);
2055                 gfs2_free_meta(dip, rgd, blk, 1);
2056                 gfs2_add_inode_blocks(&dip->i_inode, -1);
2057         }
2058 
2059         error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2060         if (error != size) {
2061                 if (error >= 0)
2062                         error = -EIO;
2063                 goto out_end_trans;
2064         }
2065 
2066         error = gfs2_meta_inode_buffer(dip, &dibh);
2067         if (error)
2068                 goto out_end_trans;
2069 
2070         gfs2_trans_add_meta(dip->i_gl, dibh);
2071         /* On the last dealloc, make this a regular file in case we crash.
2072            (We don't want to free these blocks a second time.)  */
2073         if (last_dealloc)
2074                 dip->i_inode.i_mode = S_IFREG;
2075         gfs2_dinode_out(dip, dibh->b_data);
2076         brelse(dibh);
2077 
2078 out_end_trans:
2079         gfs2_trans_end(sdp);
2080 out_rg_gunlock:
2081         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2082 out_rlist:
2083         gfs2_rlist_free(&rlist);
2084         gfs2_quota_unhold(dip);
2085 out:
2086         kvfree(ht);
2087         return error;
2088 }
2089 
2090 /**
2091  * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2092  * @dip: the directory
2093  *
2094  * Dealloc all on-disk directory leaves to FREEMETA state
2095  * Change on-disk inode type to "regular file"
2096  *
2097  * Returns: errno
2098  */
2099 
2100 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2101 {
2102         struct buffer_head *bh;
2103         struct gfs2_leaf *leaf;
2104         u32 hsize, len;
2105         u32 index = 0, next_index;
2106         __be64 *lp;
2107         u64 leaf_no;
2108         int error = 0, last;
2109 
2110         hsize = BIT(dip->i_depth);
2111 
2112         lp = gfs2_dir_get_hash_table(dip);
2113         if (IS_ERR(lp))
2114                 return PTR_ERR(lp);
2115 
2116         while (index < hsize) {
2117                 leaf_no = be64_to_cpu(lp[index]);
2118                 if (leaf_no) {
2119                         error = get_leaf(dip, leaf_no, &bh);
2120                         if (error)
2121                                 goto out;
2122                         leaf = (struct gfs2_leaf *)bh->b_data;
2123                         len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2124 
2125                         next_index = (index & ~(len - 1)) + len;
2126                         last = ((next_index >= hsize) ? 1 : 0);
2127                         error = leaf_dealloc(dip, index, len, leaf_no, bh,
2128                                              last);
2129                         brelse(bh);
2130                         if (error)
2131                                 goto out;
2132                         index = next_index;
2133                 } else
2134                         index++;
2135         }
2136 
2137         if (index != hsize) {
2138                 gfs2_consist_inode(dip);
2139                 error = -EIO;
2140         }
2141 
2142 out:
2143 
2144         return error;
2145 }
2146 
2147 /**
2148  * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2149  * @ip: the file being written to
2150  * @filname: the filename that's going to be added
2151  * @da: The structure to return dir alloc info
2152  *
2153  * Returns: 0 if ok, -ve on error
2154  */
2155 
2156 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2157                                struct gfs2_diradd *da)
2158 {
2159         struct gfs2_inode *ip = GFS2_I(inode);
2160         struct gfs2_sbd *sdp = GFS2_SB(inode);
2161         const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2162         struct gfs2_dirent *dent;
2163         struct buffer_head *bh;
2164 
2165         da->nr_blocks = 0;
2166         da->bh = NULL;
2167         da->dent = NULL;
2168 
2169         dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2170         if (!dent) {
2171                 da->nr_blocks = sdp->sd_max_dirres;
2172                 if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2173                     (GFS2_DIRENT_SIZE(name->len) < extra))
2174                         da->nr_blocks = 1;
2175                 return 0;
2176         }
2177         if (IS_ERR(dent))
2178                 return PTR_ERR(dent);
2179 
2180         if (da->save_loc) {
2181                 da->bh = bh;
2182                 da->dent = dent;
2183         } else {
2184                 brelse(bh);
2185         }
2186         return 0;
2187 }
2188 
2189 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp