mballoc.h 5.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223
  1. /*
  2. * fs/ext4/mballoc.h
  3. *
  4. * Written by: Alex Tomas <alex@clusterfs.com>
  5. *
  6. */
  7. #ifndef _EXT4_MBALLOC_H
  8. #define _EXT4_MBALLOC_H
  9. #include <linux/time.h>
  10. #include <linux/fs.h>
  11. #include <linux/namei.h>
  12. #include <linux/quotaops.h>
  13. #include <linux/buffer_head.h>
  14. #include <linux/module.h>
  15. #include <linux/swap.h>
  16. #include <linux/proc_fs.h>
  17. #include <linux/pagemap.h>
  18. #include <linux/seq_file.h>
  19. #include <linux/blkdev.h>
  20. #include <linux/mutex.h>
  21. #include "ext4_jbd2.h"
  22. #include "ext4.h"
  23. /*
  24. * with AGGRESSIVE_CHECK allocator runs consistency checks over
  25. * structures. these checks slow things down a lot
  26. */
  27. #define AGGRESSIVE_CHECK__
  28. /*
  29. * with DOUBLE_CHECK defined mballoc creates persistent in-core
  30. * bitmaps, maintains and uses them to check for double allocations
  31. */
  32. #define DOUBLE_CHECK__
  33. /*
  34. */
  35. #ifdef CONFIG_EXT4_DEBUG
  36. extern u8 mb_enable_debug;
  37. #define mb_debug(n, fmt, a...) \
  38. do { \
  39. if ((n) <= mb_enable_debug) { \
  40. printk(KERN_DEBUG "(%s, %d): %s: ", \
  41. __FILE__, __LINE__, __func__); \
  42. printk(fmt, ## a); \
  43. } \
  44. } while (0)
  45. #else
  46. #define mb_debug(n, fmt, a...)
  47. #endif
  48. #define EXT4_MB_HISTORY_ALLOC 1 /* allocation */
  49. #define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */
  50. /*
  51. * How long mballoc can look for a best extent (in found extents)
  52. */
  53. #define MB_DEFAULT_MAX_TO_SCAN 200
  54. /*
  55. * How long mballoc must look for a best extent
  56. */
  57. #define MB_DEFAULT_MIN_TO_SCAN 10
  58. /*
  59. * How many groups mballoc will scan looking for the best chunk
  60. */
  61. #define MB_DEFAULT_MAX_GROUPS_TO_SCAN 5
  62. /*
  63. * with 'ext4_mb_stats' allocator will collect stats that will be
  64. * shown at umount. The collecting costs though!
  65. */
  66. #define MB_DEFAULT_STATS 0
  67. /*
  68. * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
  69. * by the stream allocator, which purpose is to pack requests
  70. * as close each to other as possible to produce smooth I/O traffic
  71. * We use locality group prealloc space for stream request.
  72. * We can tune the same via /proc/fs/ext4/<parition>/stream_req
  73. */
  74. #define MB_DEFAULT_STREAM_THRESHOLD 16 /* 64K */
  75. /*
  76. * for which requests use 2^N search using buddies
  77. */
  78. #define MB_DEFAULT_ORDER2_REQS 2
  79. /*
  80. * default group prealloc size 512 blocks
  81. */
  82. #define MB_DEFAULT_GROUP_PREALLOC 512
  83. struct ext4_free_data {
  84. /* MUST be the first member */
  85. struct ext4_journal_cb_entry efd_jce;
  86. /* ext4_free_data private data starts from here */
  87. /* this links the free block information from group_info */
  88. struct rb_node efd_node;
  89. /* group which free block extent belongs */
  90. ext4_group_t efd_group;
  91. /* free block extent */
  92. ext4_grpblk_t efd_start_cluster;
  93. ext4_grpblk_t efd_count;
  94. /* transaction which freed this extent */
  95. tid_t efd_tid;
  96. };
  97. struct ext4_prealloc_space {
  98. struct list_head pa_inode_list;
  99. struct list_head pa_group_list;
  100. union {
  101. struct list_head pa_tmp_list;
  102. struct rcu_head pa_rcu;
  103. } u;
  104. spinlock_t pa_lock;
  105. atomic_t pa_count;
  106. unsigned pa_deleted;
  107. ext4_fsblk_t pa_pstart; /* phys. block */
  108. ext4_lblk_t pa_lstart; /* log. block */
  109. ext4_grpblk_t pa_len; /* len of preallocated chunk */
  110. ext4_grpblk_t pa_free; /* how many blocks are free */
  111. unsigned short pa_type; /* pa type. inode or group */
  112. spinlock_t *pa_obj_lock;
  113. struct inode *pa_inode; /* hack, for history only */
  114. };
  115. enum {
  116. MB_INODE_PA = 0,
  117. MB_GROUP_PA = 1
  118. };
  119. struct ext4_free_extent {
  120. ext4_lblk_t fe_logical;
  121. ext4_grpblk_t fe_start; /* In cluster units */
  122. ext4_group_t fe_group;
  123. ext4_grpblk_t fe_len; /* In cluster units */
  124. };
  125. /*
  126. * Locality group:
  127. * we try to group all related changes together
  128. * so that writeback can flush/allocate them together as well
  129. * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
  130. * (512). We store prealloc space into the hash based on the pa_free blocks
  131. * order value.ie, fls(pa_free)-1;
  132. */
  133. #define PREALLOC_TB_SIZE 10
  134. struct ext4_locality_group {
  135. /* for allocator */
  136. /* to serialize allocates */
  137. struct mutex lg_mutex;
  138. /* list of preallocations */
  139. struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
  140. spinlock_t lg_prealloc_lock;
  141. };
  142. struct ext4_allocation_context {
  143. struct inode *ac_inode;
  144. struct super_block *ac_sb;
  145. /* original request */
  146. struct ext4_free_extent ac_o_ex;
  147. /* goal request (normalized ac_o_ex) */
  148. struct ext4_free_extent ac_g_ex;
  149. /* the best found extent */
  150. struct ext4_free_extent ac_b_ex;
  151. /* copy of the best found extent taken before preallocation efforts */
  152. struct ext4_free_extent ac_f_ex;
  153. /* number of iterations done. we have to track to limit searching */
  154. unsigned long ac_ex_scanned;
  155. __u16 ac_groups_scanned;
  156. __u16 ac_found;
  157. __u16 ac_tail;
  158. __u16 ac_buddy;
  159. __u16 ac_flags; /* allocation hints */
  160. __u8 ac_status;
  161. __u8 ac_criteria;
  162. __u8 ac_2order; /* if request is to allocate 2^N blocks and
  163. * N > 0, the field stores N, otherwise 0 */
  164. __u8 ac_op; /* operation, for history only */
  165. struct page *ac_bitmap_page;
  166. struct page *ac_buddy_page;
  167. struct ext4_prealloc_space *ac_pa;
  168. struct ext4_locality_group *ac_lg;
  169. };
  170. #define AC_STATUS_CONTINUE 1
  171. #define AC_STATUS_FOUND 2
  172. #define AC_STATUS_BREAK 3
  173. struct ext4_buddy {
  174. struct page *bd_buddy_page;
  175. void *bd_buddy;
  176. struct page *bd_bitmap_page;
  177. void *bd_bitmap;
  178. struct ext4_group_info *bd_info;
  179. struct super_block *bd_sb;
  180. __u16 bd_blkbits;
  181. ext4_group_t bd_group;
  182. };
  183. static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
  184. struct ext4_free_extent *fex)
  185. {
  186. return ext4_group_first_block_no(sb, fex->fe_group) +
  187. (fex->fe_start << EXT4_SB(sb)->s_cluster_bits);
  188. }
  189. #endif