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

TOMOYO Linux Cross Reference
Linux/fs/xfs/xfs_mount.c

Version: ~ [ linux-5.2-rc1 ] ~ [ linux-5.1.2 ] ~ [ linux-5.0.16 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.43 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.119 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.176 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.179 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.139 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.67 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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 /*
  2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
  3  * All Rights Reserved.
  4  *
  5  * This program is free software; you can redistribute it and/or
  6  * modify it under the terms of the GNU General Public License as
  7  * published by the Free Software Foundation.
  8  *
  9  * This program is distributed in the hope that it would be useful,
 10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12  * GNU General Public License for more details.
 13  *
 14  * You should have received a copy of the GNU General Public License
 15  * along with this program; if not, write the Free Software Foundation,
 16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17  */
 18 #include "xfs.h"
 19 #include "xfs_fs.h"
 20 #include "xfs_types.h"
 21 #include "xfs_bit.h"
 22 #include "xfs_log.h"
 23 #include "xfs_inum.h"
 24 #include "xfs_trans.h"
 25 #include "xfs_sb.h"
 26 #include "xfs_ag.h"
 27 #include "xfs_dir2.h"
 28 #include "xfs_dmapi.h"
 29 #include "xfs_mount.h"
 30 #include "xfs_bmap_btree.h"
 31 #include "xfs_alloc_btree.h"
 32 #include "xfs_ialloc_btree.h"
 33 #include "xfs_dir2_sf.h"
 34 #include "xfs_attr_sf.h"
 35 #include "xfs_dinode.h"
 36 #include "xfs_inode.h"
 37 #include "xfs_btree.h"
 38 #include "xfs_ialloc.h"
 39 #include "xfs_alloc.h"
 40 #include "xfs_rtalloc.h"
 41 #include "xfs_bmap.h"
 42 #include "xfs_error.h"
 43 #include "xfs_rw.h"
 44 #include "xfs_quota.h"
 45 #include "xfs_fsops.h"
 46 #include "xfs_utils.h"
 47 
 48 STATIC void     xfs_unmountfs_wait(xfs_mount_t *);
 49 
 50 
 51 #ifdef HAVE_PERCPU_SB
 52 STATIC void     xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
 53                                                 int);
 54 STATIC void     xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
 55                                                 int);
 56 STATIC int      xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
 57                                                 int64_t, int);
 58 STATIC void     xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
 59 
 60 #else
 61 
 62 #define xfs_icsb_balance_counter(mp, a, b)              do { } while (0)
 63 #define xfs_icsb_balance_counter_locked(mp, a, b)       do { } while (0)
 64 #define xfs_icsb_modify_counters(mp, a, b, c)           do { } while (0)
 65 
 66 #endif
 67 
 68 static const struct {
 69         short offset;
 70         short type;     /* 0 = integer
 71                          * 1 = binary / string (no translation)
 72                          */
 73 } xfs_sb_info[] = {
 74     { offsetof(xfs_sb_t, sb_magicnum),   0 },
 75     { offsetof(xfs_sb_t, sb_blocksize),  0 },
 76     { offsetof(xfs_sb_t, sb_dblocks),    0 },
 77     { offsetof(xfs_sb_t, sb_rblocks),    0 },
 78     { offsetof(xfs_sb_t, sb_rextents),   0 },
 79     { offsetof(xfs_sb_t, sb_uuid),       1 },
 80     { offsetof(xfs_sb_t, sb_logstart),   0 },
 81     { offsetof(xfs_sb_t, sb_rootino),    0 },
 82     { offsetof(xfs_sb_t, sb_rbmino),     0 },
 83     { offsetof(xfs_sb_t, sb_rsumino),    0 },
 84     { offsetof(xfs_sb_t, sb_rextsize),   0 },
 85     { offsetof(xfs_sb_t, sb_agblocks),   0 },
 86     { offsetof(xfs_sb_t, sb_agcount),    0 },
 87     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
 88     { offsetof(xfs_sb_t, sb_logblocks),  0 },
 89     { offsetof(xfs_sb_t, sb_versionnum), 0 },
 90     { offsetof(xfs_sb_t, sb_sectsize),   0 },
 91     { offsetof(xfs_sb_t, sb_inodesize),  0 },
 92     { offsetof(xfs_sb_t, sb_inopblock),  0 },
 93     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
 94     { offsetof(xfs_sb_t, sb_blocklog),   0 },
 95     { offsetof(xfs_sb_t, sb_sectlog),    0 },
 96     { offsetof(xfs_sb_t, sb_inodelog),   0 },
 97     { offsetof(xfs_sb_t, sb_inopblog),   0 },
 98     { offsetof(xfs_sb_t, sb_agblklog),   0 },
 99     { offsetof(xfs_sb_t, sb_rextslog),   0 },
100     { offsetof(xfs_sb_t, sb_inprogress), 0 },
101     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
102     { offsetof(xfs_sb_t, sb_icount),     0 },
103     { offsetof(xfs_sb_t, sb_ifree),      0 },
104     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
105     { offsetof(xfs_sb_t, sb_frextents),  0 },
106     { offsetof(xfs_sb_t, sb_uquotino),   0 },
107     { offsetof(xfs_sb_t, sb_gquotino),   0 },
108     { offsetof(xfs_sb_t, sb_qflags),     0 },
109     { offsetof(xfs_sb_t, sb_flags),      0 },
110     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
111     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
112     { offsetof(xfs_sb_t, sb_unit),       0 },
113     { offsetof(xfs_sb_t, sb_width),      0 },
114     { offsetof(xfs_sb_t, sb_dirblklog),  0 },
115     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
116     { offsetof(xfs_sb_t, sb_logsectsize),0 },
117     { offsetof(xfs_sb_t, sb_logsunit),   0 },
118     { offsetof(xfs_sb_t, sb_features2),  0 },
119     { offsetof(xfs_sb_t, sb_bad_features2), 0 },
120     { sizeof(xfs_sb_t),                  0 }
121 };
122 
123 static DEFINE_MUTEX(xfs_uuid_table_mutex);
124 static int xfs_uuid_table_size;
125 static uuid_t *xfs_uuid_table;
126 
127 /*
128  * See if the UUID is unique among mounted XFS filesystems.
129  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
130  */
131 STATIC int
132 xfs_uuid_mount(
133         struct xfs_mount        *mp)
134 {
135         uuid_t                  *uuid = &mp->m_sb.sb_uuid;
136         int                     hole, i;
137 
138         if (mp->m_flags & XFS_MOUNT_NOUUID)
139                 return 0;
140 
141         if (uuid_is_nil(uuid)) {
142                 cmn_err(CE_WARN,
143                         "XFS: Filesystem %s has nil UUID - can't mount",
144                         mp->m_fsname);
145                 return XFS_ERROR(EINVAL);
146         }
147 
148         mutex_lock(&xfs_uuid_table_mutex);
149         for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
150                 if (uuid_is_nil(&xfs_uuid_table[i])) {
151                         hole = i;
152                         continue;
153                 }
154                 if (uuid_equal(uuid, &xfs_uuid_table[i]))
155                         goto out_duplicate;
156         }
157 
158         if (hole < 0) {
159                 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
160                         (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
161                         xfs_uuid_table_size  * sizeof(*xfs_uuid_table),
162                         KM_SLEEP);
163                 hole = xfs_uuid_table_size++;
164         }
165         xfs_uuid_table[hole] = *uuid;
166         mutex_unlock(&xfs_uuid_table_mutex);
167 
168         return 0;
169 
170  out_duplicate:
171         mutex_unlock(&xfs_uuid_table_mutex);
172         cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
173                          mp->m_fsname);
174         return XFS_ERROR(EINVAL);
175 }
176 
177 STATIC void
178 xfs_uuid_unmount(
179         struct xfs_mount        *mp)
180 {
181         uuid_t                  *uuid = &mp->m_sb.sb_uuid;
182         int                     i;
183 
184         if (mp->m_flags & XFS_MOUNT_NOUUID)
185                 return;
186 
187         mutex_lock(&xfs_uuid_table_mutex);
188         for (i = 0; i < xfs_uuid_table_size; i++) {
189                 if (uuid_is_nil(&xfs_uuid_table[i]))
190                         continue;
191                 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
192                         continue;
193                 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
194                 break;
195         }
196         ASSERT(i < xfs_uuid_table_size);
197         mutex_unlock(&xfs_uuid_table_mutex);
198 }
199 
200 
201 /*
202  * Free up the resources associated with a mount structure.  Assume that
203  * the structure was initially zeroed, so we can tell which fields got
204  * initialized.
205  */
206 STATIC void
207 xfs_free_perag(
208         xfs_mount_t     *mp)
209 {
210         if (mp->m_perag) {
211                 int     agno;
212 
213                 for (agno = 0; agno < mp->m_maxagi; agno++)
214                         if (mp->m_perag[agno].pagb_list)
215                                 kmem_free(mp->m_perag[agno].pagb_list);
216                 kmem_free(mp->m_perag);
217         }
218 }
219 
220 /*
221  * Check size of device based on the (data/realtime) block count.
222  * Note: this check is used by the growfs code as well as mount.
223  */
224 int
225 xfs_sb_validate_fsb_count(
226         xfs_sb_t        *sbp,
227         __uint64_t      nblocks)
228 {
229         ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
230         ASSERT(sbp->sb_blocklog >= BBSHIFT);
231 
232 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
233         if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
234                 return E2BIG;
235 #else                  /* Limited by UINT_MAX of sectors */
236         if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
237                 return E2BIG;
238 #endif
239         return 0;
240 }
241 
242 /*
243  * Check the validity of the SB found.
244  */
245 STATIC int
246 xfs_mount_validate_sb(
247         xfs_mount_t     *mp,
248         xfs_sb_t        *sbp,
249         int             flags)
250 {
251         /*
252          * If the log device and data device have the
253          * same device number, the log is internal.
254          * Consequently, the sb_logstart should be non-zero.  If
255          * we have a zero sb_logstart in this case, we may be trying to mount
256          * a volume filesystem in a non-volume manner.
257          */
258         if (sbp->sb_magicnum != XFS_SB_MAGIC) {
259                 xfs_fs_mount_cmn_err(flags, "bad magic number");
260                 return XFS_ERROR(EWRONGFS);
261         }
262 
263         if (!xfs_sb_good_version(sbp)) {
264                 xfs_fs_mount_cmn_err(flags, "bad version");
265                 return XFS_ERROR(EWRONGFS);
266         }
267 
268         if (unlikely(
269             sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
270                 xfs_fs_mount_cmn_err(flags,
271                         "filesystem is marked as having an external log; "
272                         "specify logdev on the\nmount command line.");
273                 return XFS_ERROR(EINVAL);
274         }
275 
276         if (unlikely(
277             sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
278                 xfs_fs_mount_cmn_err(flags,
279                         "filesystem is marked as having an internal log; "
280                         "do not specify logdev on\nthe mount command line.");
281                 return XFS_ERROR(EINVAL);
282         }
283 
284         /*
285          * More sanity checking. These were stolen directly from
286          * xfs_repair.
287          */
288         if (unlikely(
289             sbp->sb_agcount <= 0                                        ||
290             sbp->sb_sectsize < XFS_MIN_SECTORSIZE                       ||
291             sbp->sb_sectsize > XFS_MAX_SECTORSIZE                       ||
292             sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG                    ||
293             sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG                    ||
294             sbp->sb_sectsize != (1 << sbp->sb_sectlog)                  ||
295             sbp->sb_blocksize < XFS_MIN_BLOCKSIZE                       ||
296             sbp->sb_blocksize > XFS_MAX_BLOCKSIZE                       ||
297             sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG                    ||
298             sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG                    ||
299             sbp->sb_blocksize != (1 << sbp->sb_blocklog)                ||
300             sbp->sb_inodesize < XFS_DINODE_MIN_SIZE                     ||
301             sbp->sb_inodesize > XFS_DINODE_MAX_SIZE                     ||
302             sbp->sb_inodelog < XFS_DINODE_MIN_LOG                       ||
303             sbp->sb_inodelog > XFS_DINODE_MAX_LOG                       ||
304             sbp->sb_inodesize != (1 << sbp->sb_inodelog)                ||
305             (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
306             (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
307             (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
308             (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
309                 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
310                 return XFS_ERROR(EFSCORRUPTED);
311         }
312 
313         /*
314          * Sanity check AG count, size fields against data size field
315          */
316         if (unlikely(
317             sbp->sb_dblocks == 0 ||
318             sbp->sb_dblocks >
319              (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
320             sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
321                               sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
322                 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
323                 return XFS_ERROR(EFSCORRUPTED);
324         }
325 
326         /*
327          * Until this is fixed only page-sized or smaller data blocks work.
328          */
329         if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
330                 xfs_fs_mount_cmn_err(flags,
331                         "file system with blocksize %d bytes",
332                         sbp->sb_blocksize);
333                 xfs_fs_mount_cmn_err(flags,
334                         "only pagesize (%ld) or less will currently work.",
335                         PAGE_SIZE);
336                 return XFS_ERROR(ENOSYS);
337         }
338 
339         /*
340          * Currently only very few inode sizes are supported.
341          */
342         switch (sbp->sb_inodesize) {
343         case 256:
344         case 512:
345         case 1024:
346         case 2048:
347                 break;
348         default:
349                 xfs_fs_mount_cmn_err(flags,
350                         "inode size of %d bytes not supported",
351                         sbp->sb_inodesize);
352                 return XFS_ERROR(ENOSYS);
353         }
354 
355         if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
356             xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
357                 xfs_fs_mount_cmn_err(flags,
358                         "file system too large to be mounted on this system.");
359                 return XFS_ERROR(E2BIG);
360         }
361 
362         if (unlikely(sbp->sb_inprogress)) {
363                 xfs_fs_mount_cmn_err(flags, "file system busy");
364                 return XFS_ERROR(EFSCORRUPTED);
365         }
366 
367         /*
368          * Version 1 directory format has never worked on Linux.
369          */
370         if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
371                 xfs_fs_mount_cmn_err(flags,
372                         "file system using version 1 directory format");
373                 return XFS_ERROR(ENOSYS);
374         }
375 
376         return 0;
377 }
378 
379 STATIC void
380 xfs_initialize_perag_icache(
381         xfs_perag_t     *pag)
382 {
383         if (!pag->pag_ici_init) {
384                 rwlock_init(&pag->pag_ici_lock);
385                 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
386                 pag->pag_ici_init = 1;
387         }
388 }
389 
390 xfs_agnumber_t
391 xfs_initialize_perag(
392         xfs_mount_t     *mp,
393         xfs_agnumber_t  agcount)
394 {
395         xfs_agnumber_t  index, max_metadata;
396         xfs_perag_t     *pag;
397         xfs_agino_t     agino;
398         xfs_ino_t       ino;
399         xfs_sb_t        *sbp = &mp->m_sb;
400         xfs_ino_t       max_inum = XFS_MAXINUMBER_32;
401 
402         /* Check to see if the filesystem can overflow 32 bit inodes */
403         agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
404         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
405 
406         /* Clear the mount flag if no inode can overflow 32 bits
407          * on this filesystem, or if specifically requested..
408          */
409         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
410                 mp->m_flags |= XFS_MOUNT_32BITINODES;
411         } else {
412                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
413         }
414 
415         /* If we can overflow then setup the ag headers accordingly */
416         if (mp->m_flags & XFS_MOUNT_32BITINODES) {
417                 /* Calculate how much should be reserved for inodes to
418                  * meet the max inode percentage.
419                  */
420                 if (mp->m_maxicount) {
421                         __uint64_t      icount;
422 
423                         icount = sbp->sb_dblocks * sbp->sb_imax_pct;
424                         do_div(icount, 100);
425                         icount += sbp->sb_agblocks - 1;
426                         do_div(icount, sbp->sb_agblocks);
427                         max_metadata = icount;
428                 } else {
429                         max_metadata = agcount;
430                 }
431                 for (index = 0; index < agcount; index++) {
432                         ino = XFS_AGINO_TO_INO(mp, index, agino);
433                         if (ino > max_inum) {
434                                 index++;
435                                 break;
436                         }
437 
438                         /* This ag is preferred for inodes */
439                         pag = &mp->m_perag[index];
440                         pag->pagi_inodeok = 1;
441                         if (index < max_metadata)
442                                 pag->pagf_metadata = 1;
443                         xfs_initialize_perag_icache(pag);
444                 }
445         } else {
446                 /* Setup default behavior for smaller filesystems */
447                 for (index = 0; index < agcount; index++) {
448                         pag = &mp->m_perag[index];
449                         pag->pagi_inodeok = 1;
450                         xfs_initialize_perag_icache(pag);
451                 }
452         }
453         return index;
454 }
455 
456 void
457 xfs_sb_from_disk(
458         xfs_sb_t        *to,
459         xfs_dsb_t       *from)
460 {
461         to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
462         to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
463         to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
464         to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
465         to->sb_rextents = be64_to_cpu(from->sb_rextents);
466         memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
467         to->sb_logstart = be64_to_cpu(from->sb_logstart);
468         to->sb_rootino = be64_to_cpu(from->sb_rootino);
469         to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
470         to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
471         to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
472         to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
473         to->sb_agcount = be32_to_cpu(from->sb_agcount);
474         to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
475         to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
476         to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
477         to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
478         to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
479         to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
480         memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
481         to->sb_blocklog = from->sb_blocklog;
482         to->sb_sectlog = from->sb_sectlog;
483         to->sb_inodelog = from->sb_inodelog;
484         to->sb_inopblog = from->sb_inopblog;
485         to->sb_agblklog = from->sb_agblklog;
486         to->sb_rextslog = from->sb_rextslog;
487         to->sb_inprogress = from->sb_inprogress;
488         to->sb_imax_pct = from->sb_imax_pct;
489         to->sb_icount = be64_to_cpu(from->sb_icount);
490         to->sb_ifree = be64_to_cpu(from->sb_ifree);
491         to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
492         to->sb_frextents = be64_to_cpu(from->sb_frextents);
493         to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
494         to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
495         to->sb_qflags = be16_to_cpu(from->sb_qflags);
496         to->sb_flags = from->sb_flags;
497         to->sb_shared_vn = from->sb_shared_vn;
498         to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
499         to->sb_unit = be32_to_cpu(from->sb_unit);
500         to->sb_width = be32_to_cpu(from->sb_width);
501         to->sb_dirblklog = from->sb_dirblklog;
502         to->sb_logsectlog = from->sb_logsectlog;
503         to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
504         to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
505         to->sb_features2 = be32_to_cpu(from->sb_features2);
506         to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
507 }
508 
509 /*
510  * Copy in core superblock to ondisk one.
511  *
512  * The fields argument is mask of superblock fields to copy.
513  */
514 void
515 xfs_sb_to_disk(
516         xfs_dsb_t       *to,
517         xfs_sb_t        *from,
518         __int64_t       fields)
519 {
520         xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
521         xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
522         xfs_sb_field_t  f;
523         int             first;
524         int             size;
525 
526         ASSERT(fields);
527         if (!fields)
528                 return;
529 
530         while (fields) {
531                 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
532                 first = xfs_sb_info[f].offset;
533                 size = xfs_sb_info[f + 1].offset - first;
534 
535                 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
536 
537                 if (size == 1 || xfs_sb_info[f].type == 1) {
538                         memcpy(to_ptr + first, from_ptr + first, size);
539                 } else {
540                         switch (size) {
541                         case 2:
542                                 *(__be16 *)(to_ptr + first) =
543                                         cpu_to_be16(*(__u16 *)(from_ptr + first));
544                                 break;
545                         case 4:
546                                 *(__be32 *)(to_ptr + first) =
547                                         cpu_to_be32(*(__u32 *)(from_ptr + first));
548                                 break;
549                         case 8:
550                                 *(__be64 *)(to_ptr + first) =
551                                         cpu_to_be64(*(__u64 *)(from_ptr + first));
552                                 break;
553                         default:
554                                 ASSERT(0);
555                         }
556                 }
557 
558                 fields &= ~(1LL << f);
559         }
560 }
561 
562 /*
563  * xfs_readsb
564  *
565  * Does the initial read of the superblock.
566  */
567 int
568 xfs_readsb(xfs_mount_t *mp, int flags)
569 {
570         unsigned int    sector_size;
571         unsigned int    extra_flags;
572         xfs_buf_t       *bp;
573         int             error;
574 
575         ASSERT(mp->m_sb_bp == NULL);
576         ASSERT(mp->m_ddev_targp != NULL);
577 
578         /*
579          * Allocate a (locked) buffer to hold the superblock.
580          * This will be kept around at all times to optimize
581          * access to the superblock.
582          */
583         sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
584         extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
585 
586         bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
587                                 BTOBB(sector_size), extra_flags);
588         if (!bp || XFS_BUF_ISERROR(bp)) {
589                 xfs_fs_mount_cmn_err(flags, "SB read failed");
590                 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
591                 goto fail;
592         }
593         ASSERT(XFS_BUF_ISBUSY(bp));
594         ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
595 
596         /*
597          * Initialize the mount structure from the superblock.
598          * But first do some basic consistency checking.
599          */
600         xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
601 
602         error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
603         if (error) {
604                 xfs_fs_mount_cmn_err(flags, "SB validate failed");
605                 goto fail;
606         }
607 
608         /*
609          * We must be able to do sector-sized and sector-aligned IO.
610          */
611         if (sector_size > mp->m_sb.sb_sectsize) {
612                 xfs_fs_mount_cmn_err(flags,
613                         "device supports only %u byte sectors (not %u)",
614                         sector_size, mp->m_sb.sb_sectsize);
615                 error = ENOSYS;
616                 goto fail;
617         }
618 
619         /*
620          * If device sector size is smaller than the superblock size,
621          * re-read the superblock so the buffer is correctly sized.
622          */
623         if (sector_size < mp->m_sb.sb_sectsize) {
624                 XFS_BUF_UNMANAGE(bp);
625                 xfs_buf_relse(bp);
626                 sector_size = mp->m_sb.sb_sectsize;
627                 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
628                                         BTOBB(sector_size), extra_flags);
629                 if (!bp || XFS_BUF_ISERROR(bp)) {
630                         xfs_fs_mount_cmn_err(flags, "SB re-read failed");
631                         error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
632                         goto fail;
633                 }
634                 ASSERT(XFS_BUF_ISBUSY(bp));
635                 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
636         }
637 
638         /* Initialize per-cpu counters */
639         xfs_icsb_reinit_counters(mp);
640 
641         mp->m_sb_bp = bp;
642         xfs_buf_relse(bp);
643         ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
644         return 0;
645 
646  fail:
647         if (bp) {
648                 XFS_BUF_UNMANAGE(bp);
649                 xfs_buf_relse(bp);
650         }
651         return error;
652 }
653 
654 
655 /*
656  * xfs_mount_common
657  *
658  * Mount initialization code establishing various mount
659  * fields from the superblock associated with the given
660  * mount structure
661  */
662 STATIC void
663 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
664 {
665         mp->m_agfrotor = mp->m_agirotor = 0;
666         spin_lock_init(&mp->m_agirotor_lock);
667         mp->m_maxagi = mp->m_sb.sb_agcount;
668         mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
669         mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
670         mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
671         mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
672         mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
673         mp->m_blockmask = sbp->sb_blocksize - 1;
674         mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
675         mp->m_blockwmask = mp->m_blockwsize - 1;
676 
677         mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
678         mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
679         mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
680         mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
681 
682         mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
683         mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
684         mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
685         mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
686 
687         mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
688         mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
689         mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
690         mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
691 
692         mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
693         mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
694                                         sbp->sb_inopblock);
695         mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
696 }
697 
698 /*
699  * xfs_initialize_perag_data
700  *
701  * Read in each per-ag structure so we can count up the number of
702  * allocated inodes, free inodes and used filesystem blocks as this
703  * information is no longer persistent in the superblock. Once we have
704  * this information, write it into the in-core superblock structure.
705  */
706 STATIC int
707 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
708 {
709         xfs_agnumber_t  index;
710         xfs_perag_t     *pag;
711         xfs_sb_t        *sbp = &mp->m_sb;
712         uint64_t        ifree = 0;
713         uint64_t        ialloc = 0;
714         uint64_t        bfree = 0;
715         uint64_t        bfreelst = 0;
716         uint64_t        btree = 0;
717         int             error;
718 
719         for (index = 0; index < agcount; index++) {
720                 /*
721                  * read the agf, then the agi. This gets us
722                  * all the information we need and populates the
723                  * per-ag structures for us.
724                  */
725                 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
726                 if (error)
727                         return error;
728 
729                 error = xfs_ialloc_pagi_init(mp, NULL, index);
730                 if (error)
731                         return error;
732                 pag = &mp->m_perag[index];
733                 ifree += pag->pagi_freecount;
734                 ialloc += pag->pagi_count;
735                 bfree += pag->pagf_freeblks;
736                 bfreelst += pag->pagf_flcount;
737                 btree += pag->pagf_btreeblks;
738         }
739         /*
740          * Overwrite incore superblock counters with just-read data
741          */
742         spin_lock(&mp->m_sb_lock);
743         sbp->sb_ifree = ifree;
744         sbp->sb_icount = ialloc;
745         sbp->sb_fdblocks = bfree + bfreelst + btree;
746         spin_unlock(&mp->m_sb_lock);
747 
748         /* Fixup the per-cpu counters as well. */
749         xfs_icsb_reinit_counters(mp);
750 
751         return 0;
752 }
753 
754 /*
755  * Update alignment values based on mount options and sb values
756  */
757 STATIC int
758 xfs_update_alignment(xfs_mount_t *mp)
759 {
760         xfs_sb_t        *sbp = &(mp->m_sb);
761 
762         if (mp->m_dalign) {
763                 /*
764                  * If stripe unit and stripe width are not multiples
765                  * of the fs blocksize turn off alignment.
766                  */
767                 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
768                     (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
769                         if (mp->m_flags & XFS_MOUNT_RETERR) {
770                                 cmn_err(CE_WARN,
771                                         "XFS: alignment check 1 failed");
772                                 return XFS_ERROR(EINVAL);
773                         }
774                         mp->m_dalign = mp->m_swidth = 0;
775                 } else {
776                         /*
777                          * Convert the stripe unit and width to FSBs.
778                          */
779                         mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
780                         if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
781                                 if (mp->m_flags & XFS_MOUNT_RETERR) {
782                                         return XFS_ERROR(EINVAL);
783                                 }
784                                 xfs_fs_cmn_err(CE_WARN, mp,
785 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
786                                         mp->m_dalign, mp->m_swidth,
787                                         sbp->sb_agblocks);
788 
789                                 mp->m_dalign = 0;
790                                 mp->m_swidth = 0;
791                         } else if (mp->m_dalign) {
792                                 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
793                         } else {
794                                 if (mp->m_flags & XFS_MOUNT_RETERR) {
795                                         xfs_fs_cmn_err(CE_WARN, mp,
796 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
797                                                 mp->m_dalign,
798                                                 mp->m_blockmask +1);
799                                         return XFS_ERROR(EINVAL);
800                                 }
801                                 mp->m_swidth = 0;
802                         }
803                 }
804 
805                 /*
806                  * Update superblock with new values
807                  * and log changes
808                  */
809                 if (xfs_sb_version_hasdalign(sbp)) {
810                         if (sbp->sb_unit != mp->m_dalign) {
811                                 sbp->sb_unit = mp->m_dalign;
812                                 mp->m_update_flags |= XFS_SB_UNIT;
813                         }
814                         if (sbp->sb_width != mp->m_swidth) {
815                                 sbp->sb_width = mp->m_swidth;
816                                 mp->m_update_flags |= XFS_SB_WIDTH;
817                         }
818                 }
819         } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
820                     xfs_sb_version_hasdalign(&mp->m_sb)) {
821                         mp->m_dalign = sbp->sb_unit;
822                         mp->m_swidth = sbp->sb_width;
823         }
824 
825         return 0;
826 }
827 
828 /*
829  * Set the maximum inode count for this filesystem
830  */
831 STATIC void
832 xfs_set_maxicount(xfs_mount_t *mp)
833 {
834         xfs_sb_t        *sbp = &(mp->m_sb);
835         __uint64_t      icount;
836 
837         if (sbp->sb_imax_pct) {
838                 /*
839                  * Make sure the maximum inode count is a multiple
840                  * of the units we allocate inodes in.
841                  */
842                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
843                 do_div(icount, 100);
844                 do_div(icount, mp->m_ialloc_blks);
845                 mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
846                                    sbp->sb_inopblog;
847         } else {
848                 mp->m_maxicount = 0;
849         }
850 }
851 
852 /*
853  * Set the default minimum read and write sizes unless
854  * already specified in a mount option.
855  * We use smaller I/O sizes when the file system
856  * is being used for NFS service (wsync mount option).
857  */
858 STATIC void
859 xfs_set_rw_sizes(xfs_mount_t *mp)
860 {
861         xfs_sb_t        *sbp = &(mp->m_sb);
862         int             readio_log, writeio_log;
863 
864         if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
865                 if (mp->m_flags & XFS_MOUNT_WSYNC) {
866                         readio_log = XFS_WSYNC_READIO_LOG;
867                         writeio_log = XFS_WSYNC_WRITEIO_LOG;
868                 } else {
869                         readio_log = XFS_READIO_LOG_LARGE;
870                         writeio_log = XFS_WRITEIO_LOG_LARGE;
871                 }
872         } else {
873                 readio_log = mp->m_readio_log;
874                 writeio_log = mp->m_writeio_log;
875         }
876 
877         if (sbp->sb_blocklog > readio_log) {
878                 mp->m_readio_log = sbp->sb_blocklog;
879         } else {
880                 mp->m_readio_log = readio_log;
881         }
882         mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
883         if (sbp->sb_blocklog > writeio_log) {
884                 mp->m_writeio_log = sbp->sb_blocklog;
885         } else {
886                 mp->m_writeio_log = writeio_log;
887         }
888         mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
889 }
890 
891 /*
892  * Set whether we're using inode alignment.
893  */
894 STATIC void
895 xfs_set_inoalignment(xfs_mount_t *mp)
896 {
897         if (xfs_sb_version_hasalign(&mp->m_sb) &&
898             mp->m_sb.sb_inoalignmt >=
899             XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
900                 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
901         else
902                 mp->m_inoalign_mask = 0;
903         /*
904          * If we are using stripe alignment, check whether
905          * the stripe unit is a multiple of the inode alignment
906          */
907         if (mp->m_dalign && mp->m_inoalign_mask &&
908             !(mp->m_dalign & mp->m_inoalign_mask))
909                 mp->m_sinoalign = mp->m_dalign;
910         else
911                 mp->m_sinoalign = 0;
912 }
913 
914 /*
915  * Check that the data (and log if separate) are an ok size.
916  */
917 STATIC int
918 xfs_check_sizes(xfs_mount_t *mp)
919 {
920         xfs_buf_t       *bp;
921         xfs_daddr_t     d;
922         int             error;
923 
924         d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
925         if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
926                 cmn_err(CE_WARN, "XFS: size check 1 failed");
927                 return XFS_ERROR(E2BIG);
928         }
929         error = xfs_read_buf(mp, mp->m_ddev_targp,
930                              d - XFS_FSS_TO_BB(mp, 1),
931                              XFS_FSS_TO_BB(mp, 1), 0, &bp);
932         if (!error) {
933                 xfs_buf_relse(bp);
934         } else {
935                 cmn_err(CE_WARN, "XFS: size check 2 failed");
936                 if (error == ENOSPC)
937                         error = XFS_ERROR(E2BIG);
938                 return error;
939         }
940 
941         if (mp->m_logdev_targp != mp->m_ddev_targp) {
942                 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
943                 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
944                         cmn_err(CE_WARN, "XFS: size check 3 failed");
945                         return XFS_ERROR(E2BIG);
946                 }
947                 error = xfs_read_buf(mp, mp->m_logdev_targp,
948                                      d - XFS_FSB_TO_BB(mp, 1),
949                                      XFS_FSB_TO_BB(mp, 1), 0, &bp);
950                 if (!error) {
951                         xfs_buf_relse(bp);
952                 } else {
953                         cmn_err(CE_WARN, "XFS: size check 3 failed");
954                         if (error == ENOSPC)
955                                 error = XFS_ERROR(E2BIG);
956                         return error;
957                 }
958         }
959         return 0;
960 }
961 
962 /*
963  * Clear the quotaflags in memory and in the superblock.
964  */
965 int
966 xfs_mount_reset_sbqflags(
967         struct xfs_mount        *mp)
968 {
969         int                     error;
970         struct xfs_trans        *tp;
971 
972         mp->m_qflags = 0;
973 
974         /*
975          * It is OK to look at sb_qflags here in mount path,
976          * without m_sb_lock.
977          */
978         if (mp->m_sb.sb_qflags == 0)
979                 return 0;
980         spin_lock(&mp->m_sb_lock);
981         mp->m_sb.sb_qflags = 0;
982         spin_unlock(&mp->m_sb_lock);
983 
984         /*
985          * If the fs is readonly, let the incore superblock run
986          * with quotas off but don't flush the update out to disk
987          */
988         if (mp->m_flags & XFS_MOUNT_RDONLY)
989                 return 0;
990 
991 #ifdef QUOTADEBUG
992         xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
993 #endif
994 
995         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
996         error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
997                                       XFS_DEFAULT_LOG_COUNT);
998         if (error) {
999                 xfs_trans_cancel(tp, 0);
1000                 xfs_fs_cmn_err(CE_ALERT, mp,
1001                         "xfs_mount_reset_sbqflags: Superblock update failed!");
1002                 return error;
1003         }
1004 
1005         xfs_mod_sb(tp, XFS_SB_QFLAGS);
1006         return xfs_trans_commit(tp, 0);
1007 }
1008 
1009 /*
1010  * This function does the following on an initial mount of a file system:
1011  *      - reads the superblock from disk and init the mount struct
1012  *      - if we're a 32-bit kernel, do a size check on the superblock
1013  *              so we don't mount terabyte filesystems
1014  *      - init mount struct realtime fields
1015  *      - allocate inode hash table for fs
1016  *      - init directory manager
1017  *      - perform recovery and init the log manager
1018  */
1019 int
1020 xfs_mountfs(
1021         xfs_mount_t     *mp)
1022 {
1023         xfs_sb_t        *sbp = &(mp->m_sb);
1024         xfs_inode_t     *rip;
1025         __uint64_t      resblks;
1026         uint            quotamount = 0;
1027         uint            quotaflags = 0;
1028         int             error = 0;
1029 
1030         xfs_mount_common(mp, sbp);
1031 
1032         /*
1033          * Check for a mismatched features2 values.  Older kernels
1034          * read & wrote into the wrong sb offset for sb_features2
1035          * on some platforms due to xfs_sb_t not being 64bit size aligned
1036          * when sb_features2 was added, which made older superblock
1037          * reading/writing routines swap it as a 64-bit value.
1038          *
1039          * For backwards compatibility, we make both slots equal.
1040          *
1041          * If we detect a mismatched field, we OR the set bits into the
1042          * existing features2 field in case it has already been modified; we
1043          * don't want to lose any features.  We then update the bad location
1044          * with the ORed value so that older kernels will see any features2
1045          * flags, and mark the two fields as needing updates once the
1046          * transaction subsystem is online.
1047          */
1048         if (xfs_sb_has_mismatched_features2(sbp)) {
1049                 cmn_err(CE_WARN,
1050                         "XFS: correcting sb_features alignment problem");
1051                 sbp->sb_features2 |= sbp->sb_bad_features2;
1052                 sbp->sb_bad_features2 = sbp->sb_features2;
1053                 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1054 
1055                 /*
1056                  * Re-check for ATTR2 in case it was found in bad_features2
1057                  * slot.
1058                  */
1059                 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1060                    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1061                         mp->m_flags |= XFS_MOUNT_ATTR2;
1062         }
1063 
1064         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1065            (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1066                 xfs_sb_version_removeattr2(&mp->m_sb);
1067                 mp->m_update_flags |= XFS_SB_FEATURES2;
1068 
1069                 /* update sb_versionnum for the clearing of the morebits */
1070                 if (!sbp->sb_features2)
1071                         mp->m_update_flags |= XFS_SB_VERSIONNUM;
1072         }
1073 
1074         /*
1075          * Check if sb_agblocks is aligned at stripe boundary
1076          * If sb_agblocks is NOT aligned turn off m_dalign since
1077          * allocator alignment is within an ag, therefore ag has
1078          * to be aligned at stripe boundary.
1079          */
1080         error = xfs_update_alignment(mp);
1081         if (error)
1082                 goto out;
1083 
1084         xfs_alloc_compute_maxlevels(mp);
1085         xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1086         xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1087         xfs_ialloc_compute_maxlevels(mp);
1088 
1089         xfs_set_maxicount(mp);
1090 
1091         mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1092 
1093         error = xfs_uuid_mount(mp);
1094         if (error)
1095                 goto out;
1096 
1097         /*
1098          * Set the minimum read and write sizes
1099          */
1100         xfs_set_rw_sizes(mp);
1101 
1102         /*
1103          * Set the inode cluster size.
1104          * This may still be overridden by the file system
1105          * block size if it is larger than the chosen cluster size.
1106          */
1107         mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1108 
1109         /*
1110          * Set inode alignment fields
1111          */
1112         xfs_set_inoalignment(mp);
1113 
1114         /*
1115          * Check that the data (and log if separate) are an ok size.
1116          */
1117         error = xfs_check_sizes(mp);
1118         if (error)
1119                 goto out_remove_uuid;
1120 
1121         /*
1122          * Initialize realtime fields in the mount structure
1123          */
1124         error = xfs_rtmount_init(mp);
1125         if (error) {
1126                 cmn_err(CE_WARN, "XFS: RT mount failed");
1127                 goto out_remove_uuid;
1128         }
1129 
1130         /*
1131          *  Copies the low order bits of the timestamp and the randomly
1132          *  set "sequence" number out of a UUID.
1133          */
1134         uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1135 
1136         mp->m_dmevmask = 0;     /* not persistent; set after each mount */
1137 
1138         xfs_dir_mount(mp);
1139 
1140         /*
1141          * Initialize the attribute manager's entries.
1142          */
1143         mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1144 
1145         /*
1146          * Initialize the precomputed transaction reservations values.
1147          */
1148         xfs_trans_init(mp);
1149 
1150         /*
1151          * Allocate and initialize the per-ag data.
1152          */
1153         init_rwsem(&mp->m_peraglock);
1154         mp->m_perag = kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t),
1155                                   KM_MAYFAIL);
1156         if (!mp->m_perag)
1157                 goto out_remove_uuid;
1158 
1159         mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
1160 
1161         if (!sbp->sb_logblocks) {
1162                 cmn_err(CE_WARN, "XFS: no log defined");
1163                 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1164                 error = XFS_ERROR(EFSCORRUPTED);
1165                 goto out_free_perag;
1166         }
1167 
1168         /*
1169          * log's mount-time initialization. Perform 1st part recovery if needed
1170          */
1171         error = xfs_log_mount(mp, mp->m_logdev_targp,
1172                               XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1173                               XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1174         if (error) {
1175                 cmn_err(CE_WARN, "XFS: log mount failed");
1176                 goto out_free_perag;
1177         }
1178 
1179         /*
1180          * Now the log is mounted, we know if it was an unclean shutdown or
1181          * not. If it was, with the first phase of recovery has completed, we
1182          * have consistent AG blocks on disk. We have not recovered EFIs yet,
1183          * but they are recovered transactionally in the second recovery phase
1184          * later.
1185          *
1186          * Hence we can safely re-initialise incore superblock counters from
1187          * the per-ag data. These may not be correct if the filesystem was not
1188          * cleanly unmounted, so we need to wait for recovery to finish before
1189          * doing this.
1190          *
1191          * If the filesystem was cleanly unmounted, then we can trust the
1192          * values in the superblock to be correct and we don't need to do
1193          * anything here.
1194          *
1195          * If we are currently making the filesystem, the initialisation will
1196          * fail as the perag data is in an undefined state.
1197          */
1198         if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1199             !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1200              !mp->m_sb.sb_inprogress) {
1201                 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1202                 if (error)
1203                         goto out_free_perag;
1204         }
1205 
1206         /*
1207          * Get and sanity-check the root inode.
1208          * Save the pointer to it in the mount structure.
1209          */
1210         error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1211         if (error) {
1212                 cmn_err(CE_WARN, "XFS: failed to read root inode");
1213                 goto out_log_dealloc;
1214         }
1215 
1216         ASSERT(rip != NULL);
1217 
1218         if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1219                 cmn_err(CE_WARN, "XFS: corrupted root inode");
1220                 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1221                         XFS_BUFTARG_NAME(mp->m_ddev_targp),
1222                         (unsigned long long)rip->i_ino);
1223                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1224                 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1225                                  mp);
1226                 error = XFS_ERROR(EFSCORRUPTED);
1227                 goto out_rele_rip;
1228         }
1229         mp->m_rootip = rip;     /* save it */
1230 
1231         xfs_iunlock(rip, XFS_ILOCK_EXCL);
1232 
1233         /*
1234          * Initialize realtime inode pointers in the mount structure
1235          */
1236         error = xfs_rtmount_inodes(mp);
1237         if (error) {
1238                 /*
1239                  * Free up the root inode.
1240                  */
1241                 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1242                 goto out_rele_rip;
1243         }
1244 
1245         /*
1246          * If this is a read-only mount defer the superblock updates until
1247          * the next remount into writeable mode.  Otherwise we would never
1248          * perform the update e.g. for the root filesystem.
1249          */
1250         if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1251                 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1252                 if (error) {
1253                         cmn_err(CE_WARN, "XFS: failed to write sb changes");
1254                         goto out_rtunmount;
1255                 }
1256         }
1257 
1258         /*
1259          * Initialise the XFS quota management subsystem for this mount
1260          */
1261         if (XFS_IS_QUOTA_RUNNING(mp)) {
1262                 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1263                 if (error)
1264                         goto out_rtunmount;
1265         } else {
1266                 ASSERT(!XFS_IS_QUOTA_ON(mp));
1267 
1268                 /*
1269                  * If a file system had quotas running earlier, but decided to
1270                  * mount without -o uquota/pquota/gquota options, revoke the
1271                  * quotachecked license.
1272                  */
1273                 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1274                         cmn_err(CE_NOTE,
1275                                 "XFS: resetting qflags for filesystem %s",
1276                                 mp->m_fsname);
1277 
1278                         error = xfs_mount_reset_sbqflags(mp);
1279                         if (error)
1280                                 return error;
1281                 }
1282         }
1283 
1284         /*
1285          * Finish recovering the file system.  This part needed to be
1286          * delayed until after the root and real-time bitmap inodes
1287          * were consistently read in.
1288          */
1289         error = xfs_log_mount_finish(mp);
1290         if (error) {
1291                 cmn_err(CE_WARN, "XFS: log mount finish failed");
1292                 goto out_rtunmount;
1293         }
1294 
1295         /*
1296          * Complete the quota initialisation, post-log-replay component.
1297          */
1298         if (quotamount) {
1299                 ASSERT(mp->m_qflags == 0);
1300                 mp->m_qflags = quotaflags;
1301 
1302                 xfs_qm_mount_quotas(mp);
1303         }
1304 
1305 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1306         if (XFS_IS_QUOTA_ON(mp))
1307                 xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas turned on");
1308         else
1309                 xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas not turned on");
1310 #endif
1311 
1312         /*
1313          * Now we are mounted, reserve a small amount of unused space for
1314          * privileged transactions. This is needed so that transaction
1315          * space required for critical operations can dip into this pool
1316          * when at ENOSPC. This is needed for operations like create with
1317          * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1318          * are not allowed to use this reserved space.
1319          *
1320          * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1321          * This may drive us straight to ENOSPC on mount, but that implies
1322          * we were already there on the last unmount. Warn if this occurs.
1323          */
1324         resblks = mp->m_sb.sb_dblocks;
1325         do_div(resblks, 20);
1326         resblks = min_t(__uint64_t, resblks, 1024);
1327         error = xfs_reserve_blocks(mp, &resblks, NULL);
1328         if (error)
1329                 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1330                                 "Continuing without a reserve pool.");
1331 
1332         return 0;
1333 
1334  out_rtunmount:
1335         xfs_rtunmount_inodes(mp);
1336  out_rele_rip:
1337         IRELE(rip);
1338  out_log_dealloc:
1339         xfs_log_unmount(mp);
1340  out_free_perag:
1341         xfs_free_perag(mp);
1342  out_remove_uuid:
1343         xfs_uuid_unmount(mp);
1344  out:
1345         return error;
1346 }
1347 
1348 /*
1349  * This flushes out the inodes,dquots and the superblock, unmounts the
1350  * log and makes sure that incore structures are freed.
1351  */
1352 void
1353 xfs_unmountfs(
1354         struct xfs_mount        *mp)
1355 {
1356         __uint64_t              resblks;
1357         int                     error;
1358 
1359         xfs_qm_unmount_quotas(mp);
1360         xfs_rtunmount_inodes(mp);
1361         IRELE(mp->m_rootip);
1362 
1363         /*
1364          * We can potentially deadlock here if we have an inode cluster
1365          * that has been freed has its buffer still pinned in memory because
1366          * the transaction is still sitting in a iclog. The stale inodes
1367          * on that buffer will have their flush locks held until the
1368          * transaction hits the disk and the callbacks run. the inode
1369          * flush takes the flush lock unconditionally and with nothing to
1370          * push out the iclog we will never get that unlocked. hence we
1371          * need to force the log first.
1372          */
1373         xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1374         xfs_reclaim_inodes(mp, XFS_IFLUSH_ASYNC);
1375 
1376         xfs_qm_unmount(mp);
1377 
1378         /*
1379          * Flush out the log synchronously so that we know for sure
1380          * that nothing is pinned.  This is important because bflush()
1381          * will skip pinned buffers.
1382          */
1383         xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1384 
1385         xfs_binval(mp->m_ddev_targp);
1386         if (mp->m_rtdev_targp) {
1387                 xfs_binval(mp->m_rtdev_targp);
1388         }
1389 
1390         /*
1391          * Unreserve any blocks we have so that when we unmount we don't account
1392          * the reserved free space as used. This is really only necessary for
1393          * lazy superblock counting because it trusts the incore superblock
1394          * counters to be absolutely correct on clean unmount.
1395          *
1396          * We don't bother correcting this elsewhere for lazy superblock
1397          * counting because on mount of an unclean filesystem we reconstruct the
1398          * correct counter value and this is irrelevant.
1399          *
1400          * For non-lazy counter filesystems, this doesn't matter at all because
1401          * we only every apply deltas to the superblock and hence the incore
1402          * value does not matter....
1403          */
1404         resblks = 0;
1405         error = xfs_reserve_blocks(mp, &resblks, NULL);
1406         if (error)
1407                 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1408                                 "Freespace may not be correct on next mount.");
1409 
1410         error = xfs_log_sbcount(mp, 1);
1411         if (error)
1412                 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1413                                 "Freespace may not be correct on next mount.");
1414         xfs_unmountfs_writesb(mp);
1415         xfs_unmountfs_wait(mp);                 /* wait for async bufs */
1416         xfs_log_unmount_write(mp);
1417         xfs_log_unmount(mp);
1418         xfs_uuid_unmount(mp);
1419 
1420 #if defined(DEBUG)
1421         xfs_errortag_clearall(mp, 0);
1422 #endif
1423         xfs_free_perag(mp);
1424 }
1425 
1426 STATIC void
1427 xfs_unmountfs_wait(xfs_mount_t *mp)
1428 {
1429         if (mp->m_logdev_targp != mp->m_ddev_targp)
1430                 xfs_wait_buftarg(mp->m_logdev_targp);
1431         if (mp->m_rtdev_targp)
1432                 xfs_wait_buftarg(mp->m_rtdev_targp);
1433         xfs_wait_buftarg(mp->m_ddev_targp);
1434 }
1435 
1436 int
1437 xfs_fs_writable(xfs_mount_t *mp)
1438 {
1439         return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1440                 (mp->m_flags & XFS_MOUNT_RDONLY));
1441 }
1442 
1443 /*
1444  * xfs_log_sbcount
1445  *
1446  * Called either periodically to keep the on disk superblock values
1447  * roughly up to date or from unmount to make sure the values are
1448  * correct on a clean unmount.
1449  *
1450  * Note this code can be called during the process of freezing, so
1451  * we may need to use the transaction allocator which does not not
1452  * block when the transaction subsystem is in its frozen state.
1453  */
1454 int
1455 xfs_log_sbcount(
1456         xfs_mount_t     *mp,
1457         uint            sync)
1458 {
1459         xfs_trans_t     *tp;
1460         int             error;
1461 
1462         if (!xfs_fs_writable(mp))
1463                 return 0;
1464 
1465         xfs_icsb_sync_counters(mp, 0);
1466 
1467         /*
1468          * we don't need to do this if we are updating the superblock
1469          * counters on every modification.
1470          */
1471         if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1472                 return 0;
1473 
1474         tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1475         error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1476                                         XFS_DEFAULT_LOG_COUNT);
1477         if (error) {
1478                 xfs_trans_cancel(tp, 0);
1479                 return error;
1480         }
1481 
1482         xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1483         if (sync)
1484                 xfs_trans_set_sync(tp);
1485         error = xfs_trans_commit(tp, 0);
1486         return error;
1487 }
1488 
1489 int
1490 xfs_unmountfs_writesb(xfs_mount_t *mp)
1491 {
1492         xfs_buf_t       *sbp;
1493         int             error = 0;
1494 
1495         /*
1496          * skip superblock write if fs is read-only, or
1497          * if we are doing a forced umount.
1498          */
1499         if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1500                 XFS_FORCED_SHUTDOWN(mp))) {
1501 
1502                 sbp = xfs_getsb(mp, 0);
1503 
1504                 XFS_BUF_UNDONE(sbp);
1505                 XFS_BUF_UNREAD(sbp);
1506                 XFS_BUF_UNDELAYWRITE(sbp);
1507                 XFS_BUF_WRITE(sbp);
1508                 XFS_BUF_UNASYNC(sbp);
1509                 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1510                 xfsbdstrat(mp, sbp);
1511                 error = xfs_iowait(sbp);
1512                 if (error)
1513                         xfs_ioerror_alert("xfs_unmountfs_writesb",
1514                                           mp, sbp, XFS_BUF_ADDR(sbp));
1515                 xfs_buf_relse(sbp);
1516         }
1517         return error;
1518 }
1519 
1520 /*
1521  * xfs_mod_sb() can be used to copy arbitrary changes to the
1522  * in-core superblock into the superblock buffer to be logged.
1523  * It does not provide the higher level of locking that is
1524  * needed to protect the in-core superblock from concurrent
1525  * access.
1526  */
1527 void
1528 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1529 {
1530         xfs_buf_t       *bp;
1531         int             first;
1532         int             last;
1533         xfs_mount_t     *mp;
1534         xfs_sb_field_t  f;
1535 
1536         ASSERT(fields);
1537         if (!fields)
1538                 return;
1539         mp = tp->t_mountp;
1540         bp = xfs_trans_getsb(tp, mp, 0);
1541         first = sizeof(xfs_sb_t);
1542         last = 0;
1543 
1544         /* translate/copy */
1545 
1546         xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1547 
1548         /* find modified range */
1549 
1550         f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1551         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1552         first = xfs_sb_info[f].offset;
1553 
1554         f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1555         ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1556         last = xfs_sb_info[f + 1].offset - 1;
1557 
1558         xfs_trans_log_buf(tp, bp, first, last);
1559 }
1560 
1561 
1562 /*
1563  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1564  * a delta to a specified field in the in-core superblock.  Simply
1565  * switch on the field indicated and apply the delta to that field.
1566  * Fields are not allowed to dip below zero, so if the delta would
1567  * do this do not apply it and return EINVAL.
1568  *
1569  * The m_sb_lock must be held when this routine is called.
1570  */
1571 STATIC int
1572 xfs_mod_incore_sb_unlocked(
1573         xfs_mount_t     *mp,
1574         xfs_sb_field_t  field,
1575         int64_t         delta,
1576         int             rsvd)
1577 {
1578         int             scounter;       /* short counter for 32 bit fields */
1579         long long       lcounter;       /* long counter for 64 bit fields */
1580         long long       res_used, rem;
1581 
1582         /*
1583          * With the in-core superblock spin lock held, switch
1584          * on the indicated field.  Apply the delta to the
1585          * proper field.  If the fields value would dip below
1586          * 0, then do not apply the delta and return EINVAL.
1587          */
1588         switch (field) {
1589         case XFS_SBS_ICOUNT:
1590                 lcounter = (long long)mp->m_sb.sb_icount;
1591                 lcounter += delta;
1592                 if (lcounter < 0) {
1593                         ASSERT(0);
1594                         return XFS_ERROR(EINVAL);
1595                 }
1596                 mp->m_sb.sb_icount = lcounter;
1597                 return 0;
1598         case XFS_SBS_IFREE:
1599                 lcounter = (long long)mp->m_sb.sb_ifree;
1600                 lcounter += delta;
1601                 if (lcounter < 0) {
1602                         ASSERT(0);
1603                         return XFS_ERROR(EINVAL);
1604                 }
1605                 mp->m_sb.sb_ifree = lcounter;
1606                 return 0;
1607         case XFS_SBS_FDBLOCKS:
1608                 lcounter = (long long)
1609                         mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1610                 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1611 
1612                 if (delta > 0) {                /* Putting blocks back */
1613                         if (res_used > delta) {
1614                                 mp->m_resblks_avail += delta;
1615                         } else {
1616                                 rem = delta - res_used;
1617                                 mp->m_resblks_avail = mp->m_resblks;
1618                                 lcounter += rem;
1619                         }
1620                 } else {                                /* Taking blocks away */
1621 
1622                         lcounter += delta;
1623 
1624                 /*
1625                  * If were out of blocks, use any available reserved blocks if
1626                  * were allowed to.
1627                  */
1628 
1629                         if (lcounter < 0) {
1630                                 if (rsvd) {
1631                                         lcounter = (long long)mp->m_resblks_avail + delta;
1632                                         if (lcounter < 0) {
1633                                                 return XFS_ERROR(ENOSPC);
1634                                         }
1635                                         mp->m_resblks_avail = lcounter;
1636                                         return 0;
1637                                 } else {        /* not reserved */
1638                                         return XFS_ERROR(ENOSPC);
1639                                 }
1640                         }
1641                 }
1642 
1643                 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1644                 return 0;
1645         case XFS_SBS_FREXTENTS:
1646                 lcounter = (long long)mp->m_sb.sb_frextents;
1647                 lcounter += delta;
1648                 if (lcounter < 0) {
1649                         return XFS_ERROR(ENOSPC);
1650                 }
1651                 mp->m_sb.sb_frextents = lcounter;
1652                 return 0;
1653         case XFS_SBS_DBLOCKS:
1654                 lcounter = (long long)mp->m_sb.sb_dblocks;
1655                 lcounter += delta;
1656                 if (lcounter < 0) {
1657                         ASSERT(0);
1658                         return XFS_ERROR(EINVAL);
1659                 }
1660                 mp->m_sb.sb_dblocks = lcounter;
1661                 return 0;
1662         case XFS_SBS_AGCOUNT:
1663                 scounter = mp->m_sb.sb_agcount;
1664                 scounter += delta;
1665                 if (scounter < 0) {
1666                         ASSERT(0);
1667                         return XFS_ERROR(EINVAL);
1668                 }
1669                 mp->m_sb.sb_agcount = scounter;
1670                 return 0;
1671         case XFS_SBS_IMAX_PCT:
1672                 scounter = mp->m_sb.sb_imax_pct;
1673                 scounter += delta;
1674                 if (scounter < 0) {
1675                         ASSERT(0);
1676                         return XFS_ERROR(EINVAL);
1677                 }
1678                 mp->m_sb.sb_imax_pct = scounter;
1679                 return 0;
1680         case XFS_SBS_REXTSIZE:
1681                 scounter = mp->m_sb.sb_rextsize;
1682                 scounter += delta;
1683                 if (scounter < 0) {
1684                         ASSERT(0);
1685                         return XFS_ERROR(EINVAL);
1686                 }
1687                 mp->m_sb.sb_rextsize = scounter;
1688                 return 0;
1689         case XFS_SBS_RBMBLOCKS:
1690                 scounter = mp->m_sb.sb_rbmblocks;
1691                 scounter += delta;
1692                 if (scounter < 0) {
1693                         ASSERT(0);
1694                         return XFS_ERROR(EINVAL);
1695                 }
1696                 mp->m_sb.sb_rbmblocks = scounter;
1697                 return 0;
1698         case XFS_SBS_RBLOCKS:
1699                 lcounter = (long long)mp->m_sb.sb_rblocks;
1700                 lcounter += delta;
1701                 if (lcounter < 0) {
1702                         ASSERT(0);
1703                         return XFS_ERROR(EINVAL);
1704                 }
1705                 mp->m_sb.sb_rblocks = lcounter;
1706                 return 0;
1707         case XFS_SBS_REXTENTS:
1708                 lcounter = (long long)mp->m_sb.sb_rextents;
1709                 lcounter += delta;
1710                 if (lcounter < 0) {
1711                         ASSERT(0);
1712                         return XFS_ERROR(EINVAL);
1713                 }
1714                 mp->m_sb.sb_rextents = lcounter;
1715                 return 0;
1716         case XFS_SBS_REXTSLOG:
1717                 scounter = mp->m_sb.sb_rextslog;
1718                 scounter += delta;
1719                 if (scounter < 0) {
1720                         ASSERT(0);
1721                         return XFS_ERROR(EINVAL);
1722                 }
1723                 mp->m_sb.sb_rextslog = scounter;
1724                 return 0;
1725         default:
1726                 ASSERT(0);
1727                 return XFS_ERROR(EINVAL);
1728         }
1729 }
1730 
1731 /*
1732  * xfs_mod_incore_sb() is used to change a field in the in-core
1733  * superblock structure by the specified delta.  This modification
1734  * is protected by the m_sb_lock.  Just use the xfs_mod_incore_sb_unlocked()
1735  * routine to do the work.
1736  */
1737 int
1738 xfs_mod_incore_sb(
1739         xfs_mount_t     *mp,
1740         xfs_sb_field_t  field,
1741         int64_t         delta,
1742         int             rsvd)
1743 {
1744         int     status;
1745 
1746         /* check for per-cpu counters */
1747         switch (field) {
1748 #ifdef HAVE_PERCPU_SB
1749         case XFS_SBS_ICOUNT:
1750         case XFS_SBS_IFREE:
1751         case XFS_SBS_FDBLOCKS:
1752                 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1753                         status = xfs_icsb_modify_counters(mp, field,
1754                                                         delta, rsvd);
1755                         break;
1756                 }
1757                 /* FALLTHROUGH */
1758 #endif
1759         default:
1760                 spin_lock(&mp->m_sb_lock);
1761                 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1762                 spin_unlock(&mp->m_sb_lock);
1763                 break;
1764         }
1765 
1766         return status;
1767 }
1768 
1769 /*
1770  * xfs_mod_incore_sb_batch() is used to change more than one field
1771  * in the in-core superblock structure at a time.  This modification
1772  * is protected by a lock internal to this module.  The fields and
1773  * changes to those fields are specified in the array of xfs_mod_sb
1774  * structures passed in.
1775  *
1776  * Either all of the specified deltas will be applied or none of
1777  * them will.  If any modified field dips below 0, then all modifications
1778  * will be backed out and EINVAL will be returned.
1779  */
1780 int
1781 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1782 {
1783         int             status=0;
1784         xfs_mod_sb_t    *msbp;
1785 
1786         /*
1787          * Loop through the array of mod structures and apply each
1788          * individually.  If any fail, then back out all those
1789          * which have already been applied.  Do all of this within
1790          * the scope of the m_sb_lock so that all of the changes will
1791          * be atomic.
1792          */
1793         spin_lock(&mp->m_sb_lock);
1794         msbp = &msb[0];
1795         for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1796                 /*
1797                  * Apply the delta at index n.  If it fails, break
1798                  * from the loop so we'll fall into the undo loop
1799                  * below.
1800                  */
1801                 switch (msbp->msb_field) {
1802 #ifdef HAVE_PERCPU_SB
1803                 case XFS_SBS_ICOUNT:
1804                 case XFS_SBS_IFREE:
1805                 case XFS_SBS_FDBLOCKS:
1806                         if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1807                                 spin_unlock(&mp->m_sb_lock);
1808                                 status = xfs_icsb_modify_counters(mp,
1809                                                         msbp->msb_field,
1810                                                         msbp->msb_delta, rsvd);
1811                                 spin_lock(&mp->m_sb_lock);
1812                                 break;
1813                         }
1814                         /* FALLTHROUGH */
1815 #endif
1816                 default:
1817                         status = xfs_mod_incore_sb_unlocked(mp,
1818                                                 msbp->msb_field,
1819                                                 msbp->msb_delta, rsvd);
1820                         break;
1821                 }
1822 
1823                 if (status != 0) {
1824                         break;
1825                 }
1826         }
1827 
1828         /*
1829          * If we didn't complete the loop above, then back out
1830          * any changes made to the superblock.  If you add code
1831          * between the loop above and here, make sure that you
1832          * preserve the value of status. Loop back until
1833          * we step below the beginning of the array.  Make sure
1834          * we don't touch anything back there.
1835          */
1836         if (status != 0) {
1837                 msbp--;
1838                 while (msbp >= msb) {
1839                         switch (msbp->msb_field) {
1840 #ifdef HAVE_PERCPU_SB
1841                         case XFS_SBS_ICOUNT:
1842                         case XFS_SBS_IFREE:
1843                         case XFS_SBS_FDBLOCKS:
1844                                 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1845                                         spin_unlock(&mp->m_sb_lock);
1846                                         status = xfs_icsb_modify_counters(mp,
1847                                                         msbp->msb_field,
1848                                                         -(msbp->msb_delta),
1849                                                         rsvd);
1850                                         spin_lock(&mp->m_sb_lock);
1851                                         break;
1852                                 }
1853                                 /* FALLTHROUGH */
1854 #endif
1855                         default:
1856                                 status = xfs_mod_incore_sb_unlocked(mp,
1857                                                         msbp->msb_field,
1858                                                         -(msbp->msb_delta),
1859                                                         rsvd);
1860                                 break;
1861                         }
1862                         ASSERT(status == 0);
1863                         msbp--;
1864                 }
1865         }
1866         spin_unlock(&mp->m_sb_lock);
1867         return status;
1868 }
1869 
1870 /*
1871  * xfs_getsb() is called to obtain the buffer for the superblock.
1872  * The buffer is returned locked and read in from disk.
1873  * The buffer should be released with a call to xfs_brelse().
1874  *
1875  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1876  * the superblock buffer if it can be locked without sleeping.
1877  * If it can't then we'll return NULL.
1878  */
1879 xfs_buf_t *
1880 xfs_getsb(
1881         xfs_mount_t     *mp,
1882         int             flags)
1883 {
1884         xfs_buf_t       *bp;
1885 
1886         ASSERT(mp->m_sb_bp != NULL);
1887         bp = mp->m_sb_bp;
1888         if (flags & XFS_BUF_TRYLOCK) {
1889                 if (!XFS_BUF_CPSEMA(bp)) {
1890                         return NULL;
1891                 }
1892         } else {
1893                 XFS_BUF_PSEMA(bp, PRIBIO);
1894         }
1895         XFS_BUF_HOLD(bp);
1896         ASSERT(XFS_BUF_ISDONE(bp));
1897         return bp;
1898 }
1899 
1900 /*
1901  * Used to free the superblock along various error paths.
1902  */
1903 void
1904 xfs_freesb(
1905         xfs_mount_t     *mp)
1906 {
1907         xfs_buf_t       *bp;
1908 
1909         /*
1910          * Use xfs_getsb() so that the buffer will be locked
1911          * when we call xfs_buf_relse().
1912          */
1913         bp = xfs_getsb(mp, 0);
1914         XFS_BUF_UNMANAGE(bp);
1915         xfs_buf_relse(bp);
1916         mp->m_sb_bp = NULL;
1917 }
1918 
1919 /*
1920  * Used to log changes to the superblock unit and width fields which could
1921  * be altered by the mount options, as well as any potential sb_features2
1922  * fixup. Only the first superblock is updated.
1923  */
1924 int
1925 xfs_mount_log_sb(
1926         xfs_mount_t     *mp,
1927         __int64_t       fields)
1928 {
1929         xfs_trans_t     *tp;
1930         int             error;
1931 
1932         ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1933                          XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1934                          XFS_SB_VERSIONNUM));
1935 
1936         tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1937         error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1938                                 XFS_DEFAULT_LOG_COUNT);
1939         if (error) {
1940                 xfs_trans_cancel(tp, 0);
1941                 return error;
1942         }
1943         xfs_mod_sb(tp, fields);
1944         error = xfs_trans_commit(tp, 0);
1945         return error;
1946 }
1947 
1948 
1949 #ifdef HAVE_PERCPU_SB
1950 /*
1951  * Per-cpu incore superblock counters
1952  *
1953  * Simple concept, difficult implementation
1954  *
1955  * Basically, replace the incore superblock counters with a distributed per cpu
1956  * counter for contended fields (e.g.  free block count).
1957  *
1958  * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1959  * hence needs to be accurately read when we are running low on space. Hence
1960  * there is a method to enable and disable the per-cpu counters based on how
1961  * much "stuff" is available in them.
1962  *
1963  * Basically, a counter is enabled if there is enough free resource to justify
1964  * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1965  * ENOSPC), then we disable the counters to synchronise all callers and
1966  * re-distribute the available resources.
1967  *
1968  * If, once we redistributed the available resources, we still get a failure,
1969  * we disable the per-cpu counter and go through the slow path.
1970  *
1971  * The slow path is the current xfs_mod_incore_sb() function.  This means that
1972  * when we disable a per-cpu counter, we need to drain its resources back to
1973  * the global superblock. We do this after disabling the counter to prevent
1974  * more threads from queueing up on the counter.
1975  *
1976  * Essentially, this means that we still need a lock in the fast path to enable
1977  * synchronisation between the global counters and the per-cpu counters. This
1978  * is not a problem because the lock will be local to a CPU almost all the time
1979  * and have little contention except when we get to ENOSPC conditions.
1980  *
1981  * Basically, this lock becomes a barrier that enables us to lock out the fast
1982  * path while we do things like enabling and disabling counters and
1983  * synchronising the counters.
1984  *
1985  * Locking rules:
1986  *
1987  *      1. m_sb_lock before picking up per-cpu locks
1988  *      2. per-cpu locks always picked up via for_each_online_cpu() order
1989  *      3. accurate counter sync requires m_sb_lock + per cpu locks
1990  *      4. modifying per-cpu counters requires holding per-cpu lock
1991  *      5. modifying global counters requires holding m_sb_lock
1992  *      6. enabling or disabling a counter requires holding the m_sb_lock 
1993  *         and _none_ of the per-cpu locks.
1994  *
1995  * Disabled counters are only ever re-enabled by a balance operation
1996  * that results in more free resources per CPU than a given threshold.
1997  * To ensure counters don't remain disabled, they are rebalanced when
1998  * the global resource goes above a higher threshold (i.e. some hysteresis
1999  * is present to prevent thrashing).
2000  */
2001 
2002 #ifdef CONFIG_HOTPLUG_CPU
2003 /*
2004  * hot-plug CPU notifier support.
2005  *
2006  * We need a notifier per filesystem as we need to be able to identify
2007  * the filesystem to balance the counters out. This is achieved by
2008  * having a notifier block embedded in the xfs_mount_t and doing pointer
2009  * magic to get the mount pointer from the notifier block address.
2010  */
2011 STATIC int
2012 xfs_icsb_cpu_notify(
2013         struct notifier_block *nfb,
2014         unsigned long action,
2015         void *hcpu)
2016 {
2017         xfs_icsb_cnts_t *cntp;
2018         xfs_mount_t     *mp;
2019 
2020         mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2021         cntp = (xfs_icsb_cnts_t *)
2022                         per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2023         switch (action) {
2024         case CPU_UP_PREPARE:
2025         case CPU_UP_PREPARE_FROZEN:
2026                 /* Easy Case - initialize the area and locks, and
2027                  * then rebalance when online does everything else for us. */
2028                 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2029                 break;
2030         case CPU_ONLINE:
2031         case CPU_ONLINE_FROZEN:
2032                 xfs_icsb_lock(mp);
2033                 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2034                 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2035                 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2036                 xfs_icsb_unlock(mp);
2037                 break;
2038         case CPU_DEAD:
2039         case CPU_DEAD_FROZEN:
2040                 /* Disable all the counters, then fold the dead cpu's
2041                  * count into the total on the global superblock and
2042                  * re-enable the counters. */
2043                 xfs_icsb_lock(mp);
2044                 spin_lock(&mp->m_sb_lock);
2045                 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2046                 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2047                 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2048 
2049                 mp->m_sb.sb_icount += cntp->icsb_icount;
2050                 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2051                 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2052 
2053                 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2054 
2055                 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2056                 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2057                 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2058                 spin_unlock(&mp->m_sb_lock);
2059                 xfs_icsb_unlock(mp);
2060                 break;
2061         }
2062 
2063         return NOTIFY_OK;
2064 }
2065 #endif /* CONFIG_HOTPLUG_CPU */
2066 
2067 int
2068 xfs_icsb_init_counters(
2069         xfs_mount_t     *mp)
2070 {
2071         xfs_icsb_cnts_t *cntp;
2072         int             i;
2073 
2074         mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2075         if (mp->m_sb_cnts == NULL)
2076                 return -ENOMEM;
2077 
2078 #ifdef CONFIG_HOTPLUG_CPU
2079         mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2080         mp->m_icsb_notifier.priority = 0;
2081         register_hotcpu_notifier(&mp->m_icsb_notifier);
2082 #endif /* CONFIG_HOTPLUG_CPU */
2083 
2084         for_each_online_cpu(i) {
2085                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2086                 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2087         }
2088 
2089         mutex_init(&mp->m_icsb_mutex);
2090 
2091         /*
2092          * start with all counters disabled so that the
2093          * initial balance kicks us off correctly
2094          */
2095         mp->m_icsb_counters = -1;
2096         return 0;
2097 }
2098 
2099 void
2100 xfs_icsb_reinit_counters(
2101         xfs_mount_t     *mp)
2102 {
2103         xfs_icsb_lock(mp);
2104         /*
2105          * start with all counters disabled so that the
2106          * initial balance kicks us off correctly
2107          */
2108         mp->m_icsb_counters = -1;
2109         xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2110         xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2111         xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2112         xfs_icsb_unlock(mp);
2113 }
2114 
2115 void
2116 xfs_icsb_destroy_counters(
2117         xfs_mount_t     *mp)
2118 {
2119         if (mp->m_sb_cnts) {
2120                 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2121                 free_percpu(mp->m_sb_cnts);
2122         }
2123         mutex_destroy(&mp->m_icsb_mutex);
2124 }
2125 
2126 STATIC_INLINE void
2127 xfs_icsb_lock_cntr(
2128         xfs_icsb_cnts_t *icsbp)
2129 {
2130         while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2131                 ndelay(1000);
2132         }
2133 }
2134 
2135 STATIC_INLINE void
2136 xfs_icsb_unlock_cntr(
2137         xfs_icsb_cnts_t *icsbp)
2138 {
2139         clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2140 }
2141 
2142 
2143 STATIC_INLINE void
2144 xfs_icsb_lock_all_counters(
2145         xfs_mount_t     *mp)
2146 {
2147         xfs_icsb_cnts_t *cntp;
2148         int             i;
2149 
2150         for_each_online_cpu(i) {
2151                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2152                 xfs_icsb_lock_cntr(cntp);
2153         }
2154 }
2155 
2156 STATIC_INLINE void
2157 xfs_icsb_unlock_all_counters(
2158         xfs_mount_t     *mp)
2159 {
2160         xfs_icsb_cnts_t *cntp;
2161         int             i;
2162 
2163         for_each_online_cpu(i) {
2164                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2165                 xfs_icsb_unlock_cntr(cntp);
2166         }
2167 }
2168 
2169 STATIC void
2170 xfs_icsb_count(
2171         xfs_mount_t     *mp,
2172         xfs_icsb_cnts_t *cnt,
2173         int             flags)
2174 {
2175         xfs_icsb_cnts_t *cntp;
2176         int             i;
2177 
2178         memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2179 
2180         if (!(flags & XFS_ICSB_LAZY_COUNT))
2181                 xfs_icsb_lock_all_counters(mp);
2182 
2183         for_each_online_cpu(i) {
2184                 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2185                 cnt->icsb_icount += cntp->icsb_icount;
2186                 cnt->icsb_ifree += cntp->icsb_ifree;
2187                 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2188         }
2189 
2190         if (!(flags & XFS_ICSB_LAZY_COUNT))
2191                 xfs_icsb_unlock_all_counters(mp);
2192 }
2193 
2194 STATIC int
2195 xfs_icsb_counter_disabled(
2196         xfs_mount_t     *mp,
2197         xfs_sb_field_t  field)
2198 {
2199         ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2200         return test_bit(field, &mp->m_icsb_counters);
2201 }
2202 
2203 STATIC void
2204 xfs_icsb_disable_counter(
2205         xfs_mount_t     *mp,
2206         xfs_sb_field_t  field)
2207 {
2208         xfs_icsb_cnts_t cnt;
2209 
2210         ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2211 
2212         /*
2213          * If we are already disabled, then there is nothing to do
2214          * here. We check before locking all the counters to avoid
2215          * the expensive lock operation when being called in the
2216          * slow path and the counter is already disabled. This is
2217          * safe because the only time we set or clear this state is under
2218          * the m_icsb_mutex.
2219          */
2220         if (xfs_icsb_counter_disabled(mp, field))
2221                 return;
2222 
2223         xfs_icsb_lock_all_counters(mp);
2224         if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2225                 /* drain back to superblock */
2226 
2227                 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2228                 switch(field) {
2229                 case XFS_SBS_ICOUNT:
2230                         mp->m_sb.sb_icount = cnt.icsb_icount;
2231                         break;
2232                 case XFS_SBS_IFREE:
2233                         mp->m_sb.sb_ifree = cnt.icsb_ifree;
2234                         break;
2235                 case XFS_SBS_FDBLOCKS:
2236                         mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2237                         break;
2238                 default:
2239                         BUG();
2240                 }
2241         }
2242 
2243         xfs_icsb_unlock_all_counters(mp);
2244 }
2245 
2246 STATIC void
2247 xfs_icsb_enable_counter(
2248         xfs_mount_t     *mp,
2249         xfs_sb_field_t  field,
2250         uint64_t        count,
2251         uint64_t        resid)
2252 {
2253         xfs_icsb_cnts_t *cntp;
2254         int             i;
2255 
2256         ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2257 
2258         xfs_icsb_lock_all_counters(mp);
2259         for_each_online_cpu(i) {
2260                 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2261                 switch (field) {
2262                 case XFS_SBS_ICOUNT:
2263                         cntp->icsb_icount = count + resid;
2264                         break;
2265                 case XFS_SBS_IFREE:
2266                         cntp->icsb_ifree = count + resid;
2267                         break;
2268                 case XFS_SBS_FDBLOCKS:
2269                         cntp->icsb_fdblocks = count + resid;
2270                         break;
2271                 default:
2272                         BUG();
2273                         break;
2274                 }
2275                 resid = 0;
2276         }
2277         clear_bit(field, &mp->m_icsb_counters);
2278         xfs_icsb_unlock_all_counters(mp);
2279 }
2280 
2281 void
2282 xfs_icsb_sync_counters_locked(
2283         xfs_mount_t     *mp,
2284         int             flags)
2285 {
2286         xfs_icsb_cnts_t cnt;
2287 
2288         xfs_icsb_count(mp, &cnt, flags);
2289 
2290         if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2291                 mp->m_sb.sb_icount = cnt.icsb_icount;
2292         if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2293                 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2294         if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2295                 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2296 }
2297 
2298 /*
2299  * Accurate update of per-cpu counters to incore superblock
2300  */
2301 void
2302 xfs_icsb_sync_counters(
2303         xfs_mount_t     *mp,
2304         int             flags)
2305 {
2306         spin_lock(&mp->m_sb_lock);
2307         xfs_icsb_sync_counters_locked(mp, flags);
2308         spin_unlock(&mp->m_sb_lock);
2309 }
2310 
2311 /*
2312  * Balance and enable/disable counters as necessary.
2313  *
2314  * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2315  * chosen to be the same number as single on disk allocation chunk per CPU, and
2316  * free blocks is something far enough zero that we aren't going thrash when we
2317  * get near ENOSPC. We also need to supply a minimum we require per cpu to
2318  * prevent looping endlessly when xfs_alloc_space asks for more than will
2319  * be distributed to a single CPU but each CPU has enough blocks to be
2320  * reenabled.
2321  *
2322  * Note that we can be called when counters are already disabled.
2323  * xfs_icsb_disable_counter() optimises the counter locking in this case to
2324  * prevent locking every per-cpu counter needlessly.
2325  */
2326 
2327 #define XFS_ICSB_INO_CNTR_REENABLE      (uint64_t)64
2328 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2329                 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2330 STATIC void
2331 xfs_icsb_balance_counter_locked(
2332         xfs_mount_t     *mp,
2333         xfs_sb_field_t  field,
2334         int             min_per_cpu)
2335 {
2336         uint64_t        count, resid;
2337         int             weight = num_online_cpus();
2338         uint64_t        min = (uint64_t)min_per_cpu;
2339 
2340         /* disable counter and sync counter */
2341         xfs_icsb_disable_counter(mp, field);
2342 
2343         /* update counters  - first CPU gets residual*/
2344         switch (field) {
2345         case XFS_SBS_ICOUNT:
2346                 count = mp->m_sb.sb_icount;
2347                 resid = do_div(count, weight);
2348                 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2349                         return;
2350                 break;
2351         case XFS_SBS_IFREE:
2352                 count = mp->m_sb.sb_ifree;
2353                 resid = do_div(count, weight);
2354                 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2355                         return;
2356                 break;
2357         case XFS_SBS_FDBLOCKS:
2358                 count = mp->m_sb.sb_fdblocks;
2359                 resid = do_div(count, weight);
2360                 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2361                         return;
2362                 break;
2363         default:
2364                 BUG();
2365                 count = resid = 0;      /* quiet, gcc */
2366                 break;
2367         }
2368 
2369         xfs_icsb_enable_counter(mp, field, count, resid);
2370 }
2371 
2372 STATIC void
2373 xfs_icsb_balance_counter(
2374         xfs_mount_t     *mp,
2375         xfs_sb_field_t  fields,
2376         int             min_per_cpu)
2377 {
2378         spin_lock(&mp->m_sb_lock);
2379         xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2380         spin_unlock(&mp->m_sb_lock);
2381 }
2382 
2383 STATIC int
2384 xfs_icsb_modify_counters(
2385         xfs_mount_t     *mp,
2386         xfs_sb_field_t  field,
2387         int64_t         delta,
2388         int             rsvd)
2389 {
2390         xfs_icsb_cnts_t *icsbp;
2391         long long       lcounter;       /* long counter for 64 bit fields */
2392         int             cpu, ret = 0;
2393 
2394         might_sleep();
2395 again:
2396         cpu = get_cpu();
2397         icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2398 
2399         /*
2400          * if the counter is disabled, go to slow path
2401          */
2402         if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2403                 goto slow_path;
2404         xfs_icsb_lock_cntr(icsbp);
2405         if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2406                 xfs_icsb_unlock_cntr(icsbp);
2407                 goto slow_path;
2408         }
2409 
2410         switch (field) {
2411         case XFS_SBS_ICOUNT:
2412                 lcounter = icsbp->icsb_icount;
2413                 lcounter += delta;
2414                 if (unlikely(lcounter < 0))
2415                         goto balance_counter;
2416                 icsbp->icsb_icount = lcounter;
2417                 break;
2418 
2419         case XFS_SBS_IFREE:
2420                 lcounter = icsbp->icsb_ifree;
2421                 lcounter += delta;
2422                 if (unlikely(lcounter < 0))
2423                         goto balance_counter;
2424                 icsbp->icsb_ifree = lcounter;
2425                 break;
2426 
2427         case XFS_SBS_FDBLOCKS:
2428                 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2429 
2430                 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2431                 lcounter += delta;
2432                 if (unlikely(lcounter < 0))
2433                         goto balance_counter;
2434                 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2435                 break;
2436         default:
2437                 BUG();
2438                 break;
2439         }
2440         xfs_icsb_unlock_cntr(icsbp);
2441         put_cpu();
2442         return 0;
2443 
2444 slow_path:
2445         put_cpu();
2446 
2447         /*
2448          * serialise with a mutex so we don't burn lots of cpu on
2449          * the superblock lock. We still need to hold the superblock
2450          * lock, however, when we modify the global structures.
2451          */
2452         xfs_icsb_lock(mp);
2453 
2454         /*
2455          * Now running atomically.
2456          *
2457          * If the counter is enabled, someone has beaten us to rebalancing.
2458          * Drop the lock and try again in the fast path....
2459          */
2460         if (!(xfs_icsb_counter_disabled(mp, field))) {
2461                 xfs_icsb_unlock(mp);
2462                 goto again;
2463         }
2464 
2465         /*
2466          * The counter is currently disabled. Because we are
2467          * running atomically here, we know a rebalance cannot
2468          * be in progress. Hence we can go straight to operating
2469          * on the global superblock. We do not call xfs_mod_incore_sb()
2470          * here even though we need to get the m_sb_lock. Doing so
2471          * will cause us to re-enter this function and deadlock.
2472          * Hence we get the m_sb_lock ourselves and then call
2473          * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2474          * directly on the global counters.
2475          */
2476         spin_lock(&mp->m_sb_lock);
2477         ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2478         spin_unlock(&mp->m_sb_lock);
2479 
2480         /*
2481          * Now that we've modified the global superblock, we
2482          * may be able to re-enable the distributed counters
2483          * (e.g. lots of space just got freed). After that
2484          * we are done.
2485          */
2486         if (ret != ENOSPC)
2487                 xfs_icsb_balance_counter(mp, field, 0);
2488         xfs_icsb_unlock(mp);
2489         return ret;
2490 
2491 balance_counter:
2492         xfs_icsb_unlock_cntr(icsbp);
2493         put_cpu();
2494 
2495         /*
2496          * We may have multiple threads here if multiple per-cpu
2497          * counters run dry at the same time. This will mean we can
2498          * do more balances than strictly necessary but it is not
2499          * the common slowpath case.
2500          */
2501         xfs_icsb_lock(mp);
2502 
2503         /*
2504          * running atomically.
2505          *
2506          * This will leave the counter in the correct state for future
2507          * accesses. After the rebalance, we simply try again and our retry
2508          * will either succeed through the fast path or slow path without
2509          * another balance operation being required.
2510          */
2511         xfs_icsb_balance_counter(mp, field, delta);
2512         xfs_icsb_unlock(mp);
2513         goto again;
2514 }
2515 
2516 #endif
2517 

~ [ 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